diff --git a/doc/modules/ensemble.rst b/doc/modules/ensemble.rst
index 1c41763e80b80..8bff203ab062e 100644
--- a/doc/modules/ensemble.rst
+++ b/doc/modules/ensemble.rst
@@ -50,6 +50,10 @@ target values (class labels) for the training samples::
     >>> clf = RandomForestClassifier(n_estimators=10)
     >>> clf = clf.fit(X, Y)
 
+Like :ref:`decision trees <tree>`, forests of trees also extend
+to :ref:`multi-output problems <tree_multioutput>`  (if Y is an array of size
+``[n_samples, n_outputs]``).
+
 
 Random Forests
 --------------
@@ -161,6 +165,8 @@ amount of time (e.g., on large datasets).
 
  * :ref:`example_ensemble_plot_forest_iris.py`
  * :ref:`example_ensemble_plot_forest_importances_faces.py`
+ * :ref:`example_ensemble_plot_forest_multioutput.py`
+
 
 .. topic:: References
 
@@ -210,6 +216,7 @@ the matching feature to the prediction function.
  * :ref:`example_ensemble_plot_forest_importances_faces.py`
  * :ref:`example_ensemble_plot_forest_importances.py`
 
+
 .. _gradient_boosting:
 
 Gradient Tree Boosting
@@ -471,6 +478,7 @@ can be controled via the ``max_features`` parameter.
  * :ref:`example_ensemble_plot_gradient_boosting_regression.py`
  * :ref:`example_ensemble_plot_gradient_boosting_regularization.py`
 
+
 .. topic:: References
 
  .. [F2001] J. Friedman, "Greedy Function Approximation: A Gradient Boosting Machine",
diff --git a/doc/modules/tree.rst b/doc/modules/tree.rst
index 063b3a8e46fb0..a30eed4c8aff7 100644
--- a/doc/modules/tree.rst
+++ b/doc/modules/tree.rst
@@ -38,6 +38,8 @@ Some advantages of decision trees are:
       of variable. See :ref:`algorithms <tree_algorithms>` for more
       information.
 
+    - Able to handle multi-output problems.
+
     - Uses a white box model. If a given situation is observable in a model,
       the explanation for the condition is easily explained by boolean logic.
       By constrast, in a black box model (e.g., in an artificial neural
@@ -49,6 +51,7 @@ Some advantages of decision trees are:
     - Performs well even if its assumptions are somewhat violated by
       the true model from which the data were generated.
 
+
 The disadvantages of decision trees include:
 
     - Decision-tree learners can create over-complex trees that do not
@@ -78,6 +81,7 @@ The disadvantages of decision trees include:
       It is therefore recommended to balance the dataset prior to fitting
       with the decision tree.
 
+
 .. _tree_classification:
 
 Classification
@@ -87,8 +91,8 @@ Classification
 classification on a dataset.
 
 As other classifiers, :class:`DecisionTreeClassifier` take as input two
-arrays: an array X of size [n_samples, n_features] holding the training
-samples, and an array Y of integer values, size [n_samples], holding
+arrays: an array X of size ``[n_samples, n_features]`` holding the training
+samples, and an array Y of integer values, size ``[n_samples]``, holding
 the class labels for the training samples::
 
     >>> from sklearn import tree
@@ -147,6 +151,7 @@ After being fitted, the model can then be used to predict new values::
 
  * :ref:`example_tree_plot_iris.py`
 
+
 .. _tree_regression:
 
 Regression
@@ -177,6 +182,67 @@ instead of integer values::
 
  * :ref:`example_tree_plot_tree_regression.py`
 
+
+.. _tree_multioutput:
+
+Multi-output problems
+=====================
+
+A multi-output problem is a supervised learning problem with several outputs
+to predict, that is when Y is a 2d array of size ``[n_samples, n_outputs]``.
+
+When there is no correlation between the outputs, a very simple way to solve
+this kind of problem is to build n independent models, i.e. one for each
+output, and then to use those models to independently predict each one of the n
+outputs. However, because it is likely that the output values related to the
+same input are themselves correlated, an often better way is to build a single
+model capable of predicting simultaneously all n outputs. First, it requires
+lower training time since only a single estimator is built. Second, the
+generalization accuracy of the resulting estimator may often be increased.
+
+With regard to decision trees, this strategy can readily be used to support
+multi-output problems. This requires the following changes:
+
+  - Store n output values in leaves, instead of 1;
+  - Use splitting criteria that compute the average reduction across all
+    n outputs.
+
+This module offers support for multi-output problems by implementing this
+strategy in both :class:`DecisionTreeClassifier` and
+:class:`DecisionTreeRegressor`. If a decision tree is fit on an output array Y
+of size ``[n_samples, n_outputs]`` then the resulting estimator will:
+
+  * Output n_output values upon ``predict``;
+
+  * Output a list of n_output arrays of class probabilities upon
+    ``predict_proba``.
+
+
+The use of multi-output trees for regression is demonstrated in
+:ref:`example_tree_plot_tree_regression_multioutput.py`. In this example, the input
+X is a single real value and the outputs Y are the sine and cosine of X.
+
+.. figure:: ../auto_examples/tree/images/plot_tree_regression_multioutput_1.png
+   :target: ../auto_examples/tree/plot_tree_regression_multioutput.html
+   :scale: 75
+   :align: center
+
+The use of multi-output trees for classification is demonstrated in
+:ref:`example_ensemble_plot_forest_multioutput.py`. In this example, the inputs
+X are the pixels of the upper half of faces and the outputs Y are the pixels of
+the lower half of those faces.
+
+.. figure:: ../auto_examples/ensemble/images/plot_forest_multioutput_1.png
+   :target: ../auto_examples/ensemble/plot_forest_multioutput.html
+   :scale: 75
+   :align: center
+
+.. topic:: Examples:
+
+ * :ref:`example_tree_plot_tree_regression_multioutput.py`
+ * :ref:`example_ensemble_plot_forest_multioutput.py`
+
+
 .. _tree_complexity:
 
 Complexity
@@ -228,6 +294,7 @@ slowing down the algorithm significantly.
 
 Tips on practical use
 =====================
+
   * Decision trees tend to overfit on data with a large number of features.
     Getting the right ratio of samples to number of features is important, since
     a tree with few samples in high dimensional space is very likely to overfit.
@@ -259,6 +326,7 @@ Tips on practical use
   * All decision trees use Fortran ordered ``np.float32`` arrays internally.
     If training data is not in this format, a copy of the dataset will be made.
 
+
 .. _tree_algorithms:
 
 Tree algorithms: ID3, C4.5, C5.0 and CART
@@ -297,6 +365,7 @@ scikit-learn uses an optimised version of the CART algorithm.
 .. _ID3: http://en.wikipedia.org/wiki/ID3_algorithm
 .. _CART: http://en.wikipedia.org/wiki/Predictive_analytics#Classification_and_regression_trees
 
+
 .. _tree_mathematical_formulation:
 
 Mathematical formulation
diff --git a/doc/whats_new.rst b/doc/whats_new.rst
index c2d40d91c7ffe..1d8a0823dfab0 100644
--- a/doc/whats_new.rst
+++ b/doc/whats_new.rst
@@ -30,6 +30,9 @@ Changelog
 
    - A common testing framework for all estimators was added.
 
+   - Decision trees and forests of randomized trees now support multi-output
+     classification and regression problems, by `Gilles Louppe`
+
 API changes summary
 -------------------
 
diff --git a/examples/ensemble/plot_forest_importances_faces.py b/examples/ensemble/plot_forest_importances_faces.py
index 3e71d67c4f9e8..60ebaec1bb8fe 100644
--- a/examples/ensemble/plot_forest_importances_faces.py
+++ b/examples/ensemble/plot_forest_importances_faces.py
@@ -21,7 +21,7 @@
 # Number of cores to use to perform parallel fitting of the forest model
 n_jobs = 1
 
-# Loading the digits dataset
+# Load the faces dataset
 data = fetch_olivetti_faces()
 X = data.images.reshape((len(data.images), -1))
 y = data.target
diff --git a/examples/ensemble/plot_forest_multioutput.py b/examples/ensemble/plot_forest_multioutput.py
new file mode 100644
index 0000000000000..84ab43c29f5b7
--- /dev/null
+++ b/examples/ensemble/plot_forest_multioutput.py
@@ -0,0 +1,70 @@
+"""
+=========================================
+Face completion with multi-output forests
+=========================================
+
+This example shows the use of multi-output forests to complete images.
+The goal is to predict the lower half of a face given its upper half.
+
+The first row of images shows true faces. The second half illustrates
+how the forest completes the lower half of those faces.
+
+"""
+print __doc__
+
+import numpy as np
+import pylab as pl
+
+from sklearn.datasets import fetch_olivetti_faces
+from sklearn.ensemble import ExtraTreesRegressor
+from sklearn.tree import DecisionTreeRegressor
+
+
+# Load the faces datasets
+data = fetch_olivetti_faces()
+targets = data.target
+
+data = data.images.reshape((len(data.images), -1))
+train = data[targets < 30]
+test = data[targets >= 30]  # Test on independent people
+n_pixels = data.shape[1]
+
+X_train = train[:, :int(0.5 * n_pixels)]  # Upper half of the faces
+Y_train = train[:, int(0.5 * n_pixels):]  # Lower half of the faces
+X_test = test[:, :int(0.5 * n_pixels)]
+Y_test = test[:, int(0.5 * n_pixels):]
+
+# Build a multi-output forest
+forest = ExtraTreesRegressor(n_estimators=10,
+                             max_features=32,
+                             random_state=0)
+
+forest.fit(X_train, Y_train)
+Y_test_predict = forest.predict(X_test)
+
+# Plot the completed faces
+n_faces = 5
+image_shape = (64, 64)
+
+pl.figure(figsize=(2. * n_faces, 2.26 * 2))
+pl.suptitle("Face completion with multi-output forests", size=16)
+
+for i in xrange(1, 1 + n_faces):
+    face_id = np.random.randint(X_test.shape[0])
+
+    true_face = np.hstack((X_test[face_id], Y_test[face_id]))
+    completed_face = np.hstack((X_test[face_id], Y_test_predict[face_id]))
+
+    pl.subplot(2, n_faces, i)
+    pl.axis("off")
+    pl.imshow(true_face.reshape(image_shape),
+              cmap=pl.cm.gray,
+              interpolation="nearest")
+
+    pl.subplot(2, n_faces, n_faces + i)
+    pl.axis("off")
+    pl.imshow(completed_face.reshape(image_shape),
+              cmap=pl.cm.gray,
+              interpolation="nearest")
+
+pl.show()
diff --git a/examples/tree/plot_tree_regression_multioutput.py b/examples/tree/plot_tree_regression_multioutput.py
new file mode 100644
index 0000000000000..3c7d37fdb3467
--- /dev/null
+++ b/examples/tree/plot_tree_regression_multioutput.py
@@ -0,0 +1,55 @@
+"""
+===================================================================
+Multi-output Decision Tree Regression 
+===================================================================
+
+Multi-output regression with :ref:`decision trees <tree>`: the decision tree
+is used to predict simultaneously the noisy x and y observations of a circle
+given a single underlying feature. As a result, it learns local linear
+regressions approximating the circle.
+
+We can see that if the maximum depth of the tree (controlled by the
+`max_depth` parameter) is set too high, the decision trees learn too fine
+details of the training data and learn from the noise, i.e. they overfit.
+"""
+print __doc__
+
+import numpy as np
+
+# Create a random dataset
+rng = np.random.RandomState(1)
+X = np.sort(200 * rng.rand(100, 1) - 100, axis=0)
+y = np.array([np.pi * np.sin(X).ravel(), np.pi * np.cos(X).ravel()]).T
+y[::5,:] += (0.5 - rng.rand(20,2))
+
+# Fit regression model
+from sklearn.tree import DecisionTreeRegressor
+
+clf_1 = DecisionTreeRegressor(max_depth=2)
+clf_2 = DecisionTreeRegressor(max_depth=5)
+clf_3 = DecisionTreeRegressor(max_depth=8)
+clf_1.fit(X, y)
+clf_2.fit(X, y)
+clf_3.fit(X, y)
+
+# Predict
+X_test = np.arange(-100.0, 100.0, 0.01)[:, np.newaxis]
+y_1 = clf_1.predict(X_test)
+y_2 = clf_2.predict(X_test)
+y_3 = clf_3.predict(X_test)
+
+# Plot the results
+import pylab as pl
+
+pl.figure()
+pl.scatter(y[:,0], y[:,1], c="k", label="data")
+pl.scatter(y_1[:,0], y_1[:,1], c="g", label="max_depth=2")
+pl.scatter(y_2[:,0], y_2[:,1], c="r", label="max_depth=5")
+pl.scatter(y_3[:,0], y_3[:,1], c="b", label="max_depth=8")
+pl.xlim([-6, 6])
+pl.ylim([-6, 6])
+pl.xlabel("data")
+pl.ylabel("target")
+pl.title("Multi-output Decision Tree Regression")
+pl.legend()
+pl.show()
\ No newline at end of file
diff --git a/sklearn/ensemble/forest.py b/sklearn/ensemble/forest.py
index b3dfc8854b0ac..a2a400f8b551f 100644
--- a/sklearn/ensemble/forest.py
+++ b/sklearn/ensemble/forest.py
@@ -28,6 +28,8 @@ class calls the ``fit`` method of each sub-estimator on random samples
   ``ExtraTreeClassifier`` and ``ExtraTreeRegressor`` as
   sub-estimator implementations.
 
+Single and multi-output problems are both handled.
+
 """
 
 # Authors: Gilles Louppe, Brian Holt
@@ -86,19 +88,27 @@ def _parallel_build_trees(n_trees, forest, X, y,
     return trees
 
 
-def _parallel_predict_proba(trees, X, n_classes):
+def _parallel_predict_proba(trees, X, n_classes, n_outputs):
     """Private function used to compute a batch of predictions within a job."""
-    p = np.zeros((X.shape[0], n_classes))
+    n_samples = X.shape[0]
+    p = []
+
+    for k in xrange(n_outputs):
+        p.append(np.zeros((n_samples, n_classes[k])))
 
     for tree in trees:
-        if n_classes == tree.n_classes_:
-            p += tree.predict_proba(X)
+        p_tree = tree.predict_proba(X)
 
-        else:
-            proba = tree.predict_proba(X)
+        if n_outputs == 1:
+            p_tree = [p_tree]
 
-            for j, c in enumerate(tree.classes_):
-                p[:, c] += proba[:, j]
+        for k in xrange(n_outputs):
+            if n_classes[k] == tree.n_classes_[k]:
+                p[k] += p_tree[k]
+
+            else:
+                for j, c in enumerate(tree.classes_[k]):
+                    p[k][:, c] += p_tree[k][:, j]
 
     return p
 
@@ -188,7 +198,12 @@ def __init__(self, base_estimator,
         self.n_jobs = n_jobs
         self.random_state = check_random_state(random_state)
 
+        self.n_features_ = None
+        self.n_outputs_ = None
+        self.classes_ = None
+        self.n_classes_ = None
         self.feature_importances_ = None
+
         self.verbose = verbose
 
     def fit(self, X, y):
@@ -199,7 +214,7 @@ def fit(self, X, y):
         X : array-like of shape = [n_samples, n_features]
             The training input samples.
 
-        y : array-like, shape = [n_samples]
+        y : array-like, shape = [n_samples] or [n_samples, n_outputs]
             The target values (integers that correspond to classes in
             classification, real numbers in regression).
 
@@ -210,7 +225,7 @@ def fit(self, X, y):
         """
         # Precompute some data
         X = np.atleast_2d(X)
-        y = np.atleast_1d(y)
+        n_samples, self.n_features_ = X.shape
 
         if self.bootstrap:
             sample_mask = None
@@ -221,9 +236,9 @@ def fit(self, X, y):
                 raise ValueError("Out of bag estimation only available"
                         " if bootstrap=True")
 
-            sample_mask = np.ones((X.shape[0],), dtype=np.bool)
+            sample_mask = np.ones((n_samples,), dtype=np.bool)
 
-            n_jobs, _, starts = _partition_features(self, X.shape[1])
+            n_jobs, _, starts = _partition_features(self, self.n_features_)
 
             all_X_argsorted = Parallel(n_jobs=n_jobs)(
                 delayed(_parallel_X_argsort)(
@@ -232,10 +247,21 @@ def fit(self, X, y):
 
             X_argsorted = np.asfortranarray(np.hstack(all_X_argsorted))
 
+        y = np.copy(y)
+        y = np.atleast_1d(y)
+        if y.ndim == 1:
+            y = y[:, np.newaxis]
+
+        self.classes_ = []
+        self.n_classes_ = []
+        self.n_outputs_ = y.shape[1]
+
         if isinstance(self.base_estimator, ClassifierMixin):
-            self.classes_ = np.unique(y)
-            self.n_classes_ = len(self.classes_)
-            y = np.searchsorted(self.classes_, y)
+            for k in xrange(self.n_outputs_):
+                unique = np.unique(y[:, k])
+                self.classes_.append(unique)
+                self.n_classes_.append(unique.shape[0])
+                y[:, k] = np.searchsorted(unique, y[:, k])
 
         # Assign chunk of trees to jobs
         n_jobs, n_trees, _ = _partition_trees(self)
@@ -259,29 +285,66 @@ def fit(self, X, y):
         # Calculate out of bag predictions and score
         if self.oob_score:
             if isinstance(self, ClassifierMixin):
-                predictions = np.zeros((X.shape[0], self.n_classes_))
+                self.oob_decision_function_ = []
+                self.oob_score_ = 0.0
+
+                predictions = []
+                for k in xrange(self.n_outputs_):
+                    predictions.append(np.zeros((n_samples,
+                                                 self.n_classes_[k])))
+
                 for estimator in self.estimators_:
-                    mask = np.ones(X.shape[0], dtype=np.bool)
+                    mask = np.ones(n_samples, dtype=np.bool)
                     mask[estimator.indices_] = False
-                    predictions[mask, :] += estimator.predict_proba(X[mask, :])
 
-                self.oob_decision_function_ = (predictions
-                        / predictions.sum(axis=1)[:, np.newaxis])
-                self.oob_score_ = np.mean(y == np.argmax(predictions, axis=1))
+                    p_estimator = estimator.predict_proba(X[mask, :])
+                    if self.n_outputs_ == 1:
+                        p_estimator = [p_estimator]
+
+                    for k in xrange(self.n_outputs_):
+                        predictions[k][mask, :] += p_estimator[k]
+
+                for k in xrange(self.n_outputs_):
+                    decision = predictions[k] \
+                               / predictions[k].sum(axis=1)[:, np.newaxis]
+                    self.oob_decision_function_.append(decision)
+
+                    self.oob_score_ += np.mean(y[:, k] \
+                                       == np.argmax(predictions[k], axis=1))
+
+                if self.n_outputs_ == 1:
+                    self.oob_decision_function_ = \
+                        self.oob_decision_function_[0]
+
+                self.oob_score_ /= self.n_outputs_
 
             else:
                 # Regression:
-                predictions = np.zeros(X.shape[0])
-                n_predictions = np.zeros(X.shape[0])
+                predictions = np.zeros((n_samples, self.n_outputs_))
+                n_predictions = np.zeros((n_samples, self.n_outputs_))
+
                 for estimator in self.estimators_:
-                    mask = np.ones(X.shape[0], dtype=np.bool)
+                    mask = np.ones(n_samples, dtype=np.bool)
                     mask[estimator.indices_] = False
-                    predictions[mask] += estimator.predict(X[mask, :])
-                    n_predictions[mask] += 1
+
+                    p_estimator = estimator.predict(X[mask, :])
+                    if self.n_outputs_ == 1:
+                        p_estimator = p_estimator[:, np.newaxis]
+
+                    predictions[mask, :] += p_estimator
+                    n_predictions[mask, :] += 1
+
                 predictions /= n_predictions
 
                 self.oob_prediction_ = predictions
-                self.oob_score_ = r2_score(y, predictions)
+                if self.n_outputs_ == 1:
+                    self.oob_prediction_ = \
+                        self.oob_prediction_.reshape((n_samples, ))
+
+                self.oob_score_ = 0.0
+                for k in xrange(self.n_outputs_):
+                    self.oob_score_ += r2_score(y[:, k], predictions[:, k])
+                self.oob_score_ /= self.n_outputs_
 
         # Sum the importances
         if self.compute_importances:
@@ -334,11 +397,25 @@ def predict(self, X):
 
         Returns
         -------
-        y : array of shape = [n_samples]
+        y : array of shape = [n_samples] or [n_samples, n_outputs]
             The predicted classes.
         """
-        return self.classes_.take(
-            np.argmax(self.predict_proba(X), axis=1),  axis=0)
+        n_samples = len(X)
+
+        P = self.predict_proba(X)
+        if self.n_outputs_ == 1:
+            P = [P]
+
+        predictions = np.zeros((n_samples, self.n_outputs_))
+
+        for k in xrange(self.n_outputs_):
+            predictions[:, k] = self.classes_[k].take(np.argmax(P[k], axis=1),
+                                                      axis=0)
+
+        if self.n_outputs_ == 1:
+            predictions = predictions.reshape((n_samples, ))
+
+        return predictions
 
     def predict_proba(self, X):
         """Predict class probabilities for X.
@@ -353,7 +430,8 @@ def predict_proba(self, X):
 
         Returns
         -------
-        p : array of shape = [n_samples]
+        p : array of shape = [n_samples, n_classes], or a list of n_outputs
+            such arrays if n_outputs > 1.
             The class probabilities of the input samples. Classes are
             ordered by arithmetical order.
         """
@@ -367,13 +445,26 @@ def predict_proba(self, X):
         all_p = Parallel(n_jobs=n_jobs)(
             delayed(_parallel_predict_proba)(
                 self.estimators_[starts[i]:starts[i + 1]],
-                X, self.n_classes_)
+                X,
+                self.n_classes_,
+                self.n_outputs_)
             for i in xrange(n_jobs))
 
         # Reduce
-        p = sum(all_p) / self.n_estimators
+        p = all_p[0]
+
+        for j in xrange(1, self.n_jobs):
+            for k in xrange(self.n_outputs_):
+                p[k] += all_p[j][k]
+
+        for k in xrange(self.n_outputs_):
+            p[k] /= self.n_estimators
 
-        return p
+        if self.n_outputs_ == 1:
+            return p[0]
+
+        else:
+            return p
 
     def predict_log_proba(self, X):
         """Predict class log-probabilities for X.
@@ -388,11 +479,21 @@ def predict_log_proba(self, X):
 
         Returns
         -------
-        p : array of shape = [n_samples]
+        p : array of shape = [n_samples, n_classes], or a list of n_outputs
+            such arrays if n_outputs > 1.
             The class log-probabilities of the input samples. Classes are
             ordered by arithmetical order.
         """
-        return np.log(self.predict_proba(X))
+        proba = self.predict_proba(X)
+
+        if self.n_outputs_ == 1:
+            return np.log(proba)
+
+        else:
+            for k in xrange(self.n_outputs_):
+                proba[k] = np.log(proba[k])
+
+            return proba
 
 
 class ForestRegressor(BaseForest, RegressorMixin):
@@ -437,7 +538,7 @@ def predict(self, X):
 
         Returns
         -------
-        y: array of shape = [n_samples]
+        y: array of shape = [n_samples] or [n_samples, n_outputs]
             The predicted values.
         """
         # Check data
diff --git a/sklearn/ensemble/gradient_boosting.py b/sklearn/ensemble/gradient_boosting.py
index dcaee074c37a9..0579f3da5134f 100644
--- a/sklearn/ensemble/gradient_boosting.py
+++ b/sklearn/ensemble/gradient_boosting.py
@@ -177,7 +177,7 @@ def update_terminal_regions(self, tree, X, y, residual, y_pred,
                                          y_pred[:, k])
 
         # update predictions (both in-bag and out-of-bag)
-        y_pred[:, k] += learn_rate * tree.value[:, 0].take(terminal_regions,
+        y_pred[:, k] += learn_rate * tree.value[:, 0, 0].take(terminal_regions,
                                                            axis=0)
 
     @abstractmethod
@@ -239,8 +239,8 @@ def _update_terminal_region(self, tree, terminal_regions, leaf, X, y,
                                 residual, pred):
         """LAD updates terminal regions to median estimates. """
         terminal_region = np.where(terminal_regions == leaf)[0]
-        tree.value[leaf, 0] = np.median(y.take(terminal_region, axis=0) - \
-                                        pred.take(terminal_region, axis=0))
+        tree.value[leaf, 0, 0] = np.median(y.take(terminal_region, axis=0) - \
+                                           pred.take(terminal_region, axis=0))
 
 
 class HuberLossFunction(RegressionLossFunction):
@@ -364,9 +364,9 @@ def _update_terminal_region(self, tree, terminal_regions, leaf, X, y,
         denominator = np.sum((y - residual) * (1 - y + residual))
 
         if denominator == 0.0:
-            tree.value[leaf, 0] = 0.0
+            tree.value[leaf, 0, 0] = 0.0
         else:
-            tree.value[leaf, 0] = numerator / denominator
+            tree.value[leaf, 0, 0] = numerator / denominator
 
 
 class MultinomialDeviance(LossFunction):
@@ -414,9 +414,9 @@ def _update_terminal_region(self, tree, terminal_regions, leaf, X, y,
         denominator = np.sum((y - residual) * (1.0 - y + residual))
 
         if denominator == 0.0:
-            tree.value[leaf, 0] = 0.0
+            tree.value[leaf, 0, 0] = 0.0
         else:
-            tree.value[leaf, 0] = numerator / denominator
+            tree.value[leaf, 0, 0] = numerator / denominator
 
 
 LOSS_FUNCTIONS = {'ls': LeastSquaresError,
@@ -491,8 +491,8 @@ def fit_stage(self, i, X, X_argsorted, y, y_pred, sample_mask):
             residual = loss.negative_gradient(y, y_pred, k=k)
 
             # induce regression tree on residuals
-            tree = Tree(1, self.n_features)
-            tree.build(X, residual, MSE(), self.max_depth,
+            tree = Tree(1, self.n_features, 1)
+            tree.build(X, residual[:, np.newaxis], MSE(1), self.max_depth,
                        self.min_samples_split, self.min_samples_leaf, 0.0,
                        self.max_features, self.random_state, _find_best_split,
                        sample_mask, X_argsorted)
diff --git a/sklearn/ensemble/tests/test_forest.py b/sklearn/ensemble/tests/test_forest.py
index 0259c289273cf..3aed8c73fb96f 100644
--- a/sklearn/ensemble/tests/test_forest.py
+++ b/sklearn/ensemble/tests/test_forest.py
@@ -260,7 +260,7 @@ def test_pickle():
     import pickle
 
     # Random forest
-    obj = RandomForestClassifier()
+    obj = RandomForestClassifier(random_state=0)
     obj.fit(iris.data, iris.target)
     score = obj.score(iris.data, iris.target)
     s = pickle.dumps(obj)
@@ -270,7 +270,7 @@ def test_pickle():
     score2 = obj2.score(iris.data, iris.target)
     assert_true(score == score2)
 
-    obj = RandomForestRegressor()
+    obj = RandomForestRegressor(random_state=0)
     obj.fit(boston.data, boston.target)
     score = obj.score(boston.data, boston.target)
     s = pickle.dumps(obj)
@@ -281,7 +281,7 @@ def test_pickle():
     assert_true(score == score2)
 
     # Extra-trees
-    obj = ExtraTreesClassifier()
+    obj = ExtraTreesClassifier(random_state=0)
     obj.fit(iris.data, iris.target)
     score = obj.score(iris.data, iris.target)
     s = pickle.dumps(obj)
@@ -291,7 +291,7 @@ def test_pickle():
     score2 = obj2.score(iris.data, iris.target)
     assert_true(score == score2)
 
-    obj = ExtraTreesRegressor()
+    obj = ExtraTreesRegressor(random_state=0)
     obj.fit(boston.data, boston.target)
     score = obj.score(boston.data, boston.target)
     s = pickle.dumps(obj)
@@ -302,6 +302,61 @@ def test_pickle():
     assert_true(score == score2)
 
 
+def test_multioutput():
+    """Check estimators on multi-output problems."""
+
+    X = [[-2, -1],
+         [-1, -1],
+         [-1, -2],
+         [1, 1],
+         [1, 2],
+         [2, 1],
+         [-2, 1],
+         [-1, 1],
+         [-1, 2],
+         [2, -1],
+         [1, -1],
+         [1, -2]]
+
+    y = [[-1, 0],
+         [-1, 0],
+         [-1, 0],
+         [1, 1],
+         [1, 1],
+         [1, 1],
+         [-1, 2],
+         [-1, 2],
+         [-1, 2],
+         [1, 3],
+         [1, 3],
+         [1, 3]]
+
+    T = [[-1, -1], [1, 1], [-1, 1], [1, -1]]
+    y_true = [[-1, 0], [1, 1], [-1, 2], [1, 3]]
+
+    # toy classification problem
+    clf = ExtraTreesClassifier(random_state=0)
+    y_hat = clf.fit(X, y).predict(T)
+    assert_array_equal(y_hat, y_true)
+    assert_equal(y_hat.shape, (4, 2))
+
+    proba = clf.predict_proba(T)
+    assert_equal(len(proba), 2)
+    assert_equal(proba[0].shape, (4, 2))
+    assert_equal(proba[1].shape, (4, 4))
+
+    log_proba = clf.predict_log_proba(T)
+    assert_equal(len(log_proba), 2)
+    assert_equal(log_proba[0].shape, (4, 2))
+    assert_equal(log_proba[1].shape, (4, 4))
+
+    # toy regression problem
+    clf = ExtraTreesRegressor(random_state=5)
+    y_hat = clf.fit(X, y).predict(T)
+    assert_almost_equal(y_hat, y_true)
+    assert_equal(y_hat.shape, (4, 2))
+
+
 if __name__ == "__main__":
     import nose
     nose.runmodule()
diff --git a/sklearn/tree/_tree.c b/sklearn/tree/_tree.c
index ddc9c6a2d0c6b..02359faf1f29d 100644
--- a/sklearn/tree/_tree.c
+++ b/sklearn/tree/_tree.c
@@ -1,4 +1,4 @@
-/* Generated by Cython 0.16 on Tue Jun 26 14:21:54 2012 */
+/* Generated by Cython 0.16 on Mon Jul  9 17:29:13 2012 */
 
 #define PY_SSIZE_T_CLEAN
 #include "Python.h"
@@ -621,7 +621,7 @@ typedef __pyx_t_5numpy_float32_t __pyx_t_7sklearn_4tree_5_tree_DTYPE_t;
  * ctypedef np.float32_t DTYPE_t
  * ctypedef np.int8_t BOOL_t             # <<<<<<<<<<<<<<
  * 
- * cdef extern from "math.h":
+ * cdef extern from "stdlib.h":
  */
 typedef __pyx_t_5numpy_int8_t __pyx_t_7sklearn_4tree_5_tree_BOOL_t;
 #if CYTHON_CCOMPLEX
@@ -689,7 +689,7 @@ typedef npy_clongdouble __pyx_t_5numpy_clongdouble_t;
  */
 typedef npy_cdouble __pyx_t_5numpy_complex_t;
 
-/* "sklearn/tree/_tree.pyx":40
+/* "sklearn/tree/_tree.pyx":47
  * #    be the proportion of class k observations in node m
  * 
  * cdef class Criterion:             # <<<<<<<<<<<<<<
@@ -702,7 +702,7 @@ struct __pyx_obj_7sklearn_4tree_5_tree_Criterion {
 };
 
 
-/* "sklearn/tree/_tree.pyx":67
+/* "sklearn/tree/_tree.pyx":81
  * 
  * 
  * cdef class ClassificationCriterion(Criterion):             # <<<<<<<<<<<<<<
@@ -711,21 +711,20 @@ struct __pyx_obj_7sklearn_4tree_5_tree_Criterion {
  */
 struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion {
   struct __pyx_obj_7sklearn_4tree_5_tree_Criterion __pyx_base;
-  int n_classes;
+  int n_outputs;
+  int *n_classes;
   int n_samples;
+  int label_count_stride;
   int *label_count_left;
   int *label_count_right;
   int *label_count_init;
   int n_left;
   int n_right;
-  PyObject *ndarray_label_count_left;
-  PyObject *ndarray_label_count_right;
-  PyObject *ndarray_label_count_init;
 };
 
 
-/* "sklearn/tree/_tree.pyx":181
- * 
+/* "sklearn/tree/_tree.pyx":255
+ *         return value
  * 
  * cdef class Gini(ClassificationCriterion):             # <<<<<<<<<<<<<<
  *     """Gini Index splitting criteria.
@@ -736,7 +735,7 @@ struct __pyx_obj_7sklearn_4tree_5_tree_Gini {
 };
 
 
-/* "sklearn/tree/_tree.pyx":217
+/* "sklearn/tree/_tree.pyx":306
  * 
  * 
  * cdef class Entropy(ClassificationCriterion):             # <<<<<<<<<<<<<<
@@ -748,7 +747,7 @@ struct __pyx_obj_7sklearn_4tree_5_tree_Entropy {
 };
 
 
-/* "sklearn/tree/_tree.pyx":245
+/* "sklearn/tree/_tree.pyx":348
  * 
  * 
  * cdef class RegressionCriterion(Criterion):             # <<<<<<<<<<<<<<
@@ -757,21 +756,22 @@ struct __pyx_obj_7sklearn_4tree_5_tree_Entropy {
  */
 struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion {
   struct __pyx_obj_7sklearn_4tree_5_tree_Criterion __pyx_base;
+  int n_outputs;
   int n_samples;
+  double *mean_left;
+  double *mean_right;
+  double *mean_init;
+  double *sq_sum_left;
+  double *sq_sum_right;
+  double *sq_sum_init;
+  double *var_left;
+  double *var_right;
   int n_right;
   int n_left;
-  double mean_left;
-  double mean_right;
-  double mean_init;
-  double sq_sum_right;
-  double sq_sum_left;
-  double sq_sum_init;
-  double var_left;
-  double var_right;
 };
 
 
-/* "sklearn/tree/_tree.pyx":394
+/* "sklearn/tree/_tree.pyx":576
  * 
  * 
  * cdef class MSE(RegressionCriterion):             # <<<<<<<<<<<<<<
@@ -784,7 +784,7 @@ struct __pyx_obj_7sklearn_4tree_5_tree_MSE {
 
 
 
-/* "sklearn/tree/_tree.pyx":40
+/* "sklearn/tree/_tree.pyx":47
  * #    be the proportion of class k observations in node m
  * 
  * cdef class Criterion:             # <<<<<<<<<<<<<<
@@ -793,16 +793,16 @@ struct __pyx_obj_7sklearn_4tree_5_tree_MSE {
  */
 
 struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion {
-  void (*init)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *, int, int);
+  void (*init)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *, int, int);
   void (*reset)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *);
-  int (*update)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int, int, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *);
+  int (*update)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int, int, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int, int *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *);
   double (*eval)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *);
   PyArrayObject *(*init_value)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int __pyx_skip_dispatch);
 };
 static struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *__pyx_vtabptr_7sklearn_4tree_5_tree_Criterion;
 
 
-/* "sklearn/tree/_tree.pyx":245
+/* "sklearn/tree/_tree.pyx":348
  * 
  * 
  * cdef class RegressionCriterion(Criterion):             # <<<<<<<<<<<<<<
@@ -816,7 +816,7 @@ struct __pyx_vtabstruct_7sklearn_4tree_5_tree_RegressionCriterion {
 static struct __pyx_vtabstruct_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_vtabptr_7sklearn_4tree_5_tree_RegressionCriterion;
 
 
-/* "sklearn/tree/_tree.pyx":394
+/* "sklearn/tree/_tree.pyx":576
  * 
  * 
  * cdef class MSE(RegressionCriterion):             # <<<<<<<<<<<<<<
@@ -830,7 +830,7 @@ struct __pyx_vtabstruct_7sklearn_4tree_5_tree_MSE {
 static struct __pyx_vtabstruct_7sklearn_4tree_5_tree_MSE *__pyx_vtabptr_7sklearn_4tree_5_tree_MSE;
 
 
-/* "sklearn/tree/_tree.pyx":67
+/* "sklearn/tree/_tree.pyx":81
  * 
  * 
  * cdef class ClassificationCriterion(Criterion):             # <<<<<<<<<<<<<<
@@ -844,8 +844,8 @@ struct __pyx_vtabstruct_7sklearn_4tree_5_tree_ClassificationCriterion {
 static struct __pyx_vtabstruct_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_vtabptr_7sklearn_4tree_5_tree_ClassificationCriterion;
 
 
-/* "sklearn/tree/_tree.pyx":181
- * 
+/* "sklearn/tree/_tree.pyx":255
+ *         return value
  * 
  * cdef class Gini(ClassificationCriterion):             # <<<<<<<<<<<<<<
  *     """Gini Index splitting criteria.
@@ -858,7 +858,7 @@ struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Gini {
 static struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Gini *__pyx_vtabptr_7sklearn_4tree_5_tree_Gini;
 
 
-/* "sklearn/tree/_tree.pyx":217
+/* "sklearn/tree/_tree.pyx":306
  * 
  * 
  * cdef class Entropy(ClassificationCriterion):             # <<<<<<<<<<<<<<
@@ -924,34 +924,105 @@ static struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Entropy *__pyx_vtabptr_7skl
 #define __Pyx_CLEAR(r)    do { PyObject* tmp = ((PyObject*)(r)); r = NULL; __Pyx_DECREF(tmp);} while(0)
 #define __Pyx_XCLEAR(r)   do { if((r) != NULL) {PyObject* tmp = ((PyObject*)(r)); r = NULL; __Pyx_DECREF(tmp);}} while(0)
 
-static PyObject *__Pyx_GetName(PyObject *dict, PyObject *name); /*proto*/
-
 static CYTHON_INLINE int __Pyx_TypeTest(PyObject *obj, PyTypeObject *type); /*proto*/
 
+static void __Pyx_RaiseArgtupleInvalid(const char* func_name, int exact,
+    Py_ssize_t num_min, Py_ssize_t num_max, Py_ssize_t num_found); /*proto*/
+
 static void __Pyx_RaiseDoubleKeywordsError(const char* func_name, PyObject* kw_name); /*proto*/
 
 static int __Pyx_ParseOptionalKeywords(PyObject *kwds, PyObject **argnames[], \
     PyObject *kwds2, PyObject *values[], Py_ssize_t num_pos_args, \
     const char* function_name); /*proto*/
 
-static void __Pyx_RaiseArgtupleInvalid(const char* func_name, int exact,
-    Py_ssize_t num_min, Py_ssize_t num_max, Py_ssize_t num_found); /*proto*/
+static CYTHON_INLINE PyObject *__Pyx_GetItemInt_Generic(PyObject *o, PyObject* j) {
+    PyObject *r;
+    if (!j) return NULL;
+    r = PyObject_GetItem(o, j);
+    Py_DECREF(j);
+    return r;
+}
+#define __Pyx_GetItemInt_List(o, i, size, to_py_func) (((size) <= sizeof(Py_ssize_t)) ? \
+                                                    __Pyx_GetItemInt_List_Fast(o, i) : \
+                                                    __Pyx_GetItemInt_Generic(o, to_py_func(i)))
+static CYTHON_INLINE PyObject *__Pyx_GetItemInt_List_Fast(PyObject *o, Py_ssize_t i) {
+    if (likely(o != Py_None)) {
+        if (likely((0 <= i) & (i < PyList_GET_SIZE(o)))) {
+            PyObject *r = PyList_GET_ITEM(o, i);
+            Py_INCREF(r);
+            return r;
+        }
+        else if ((-PyList_GET_SIZE(o) <= i) & (i < 0)) {
+            PyObject *r = PyList_GET_ITEM(o, PyList_GET_SIZE(o) + i);
+            Py_INCREF(r);
+            return r;
+        }
+    }
+    return __Pyx_GetItemInt_Generic(o, PyInt_FromSsize_t(i));
+}
+#define __Pyx_GetItemInt_Tuple(o, i, size, to_py_func) (((size) <= sizeof(Py_ssize_t)) ? \
+                                                    __Pyx_GetItemInt_Tuple_Fast(o, i) : \
+                                                    __Pyx_GetItemInt_Generic(o, to_py_func(i)))
+static CYTHON_INLINE PyObject *__Pyx_GetItemInt_Tuple_Fast(PyObject *o, Py_ssize_t i) {
+    if (likely(o != Py_None)) {
+        if (likely((0 <= i) & (i < PyTuple_GET_SIZE(o)))) {
+            PyObject *r = PyTuple_GET_ITEM(o, i);
+            Py_INCREF(r);
+            return r;
+        }
+        else if ((-PyTuple_GET_SIZE(o) <= i) & (i < 0)) {
+            PyObject *r = PyTuple_GET_ITEM(o, PyTuple_GET_SIZE(o) + i);
+            Py_INCREF(r);
+            return r;
+        }
+    }
+    return __Pyx_GetItemInt_Generic(o, PyInt_FromSsize_t(i));
+}
+#define __Pyx_GetItemInt(o, i, size, to_py_func) (((size) <= sizeof(Py_ssize_t)) ? \
+                                                    __Pyx_GetItemInt_Fast(o, i) : \
+                                                    __Pyx_GetItemInt_Generic(o, to_py_func(i)))
+static CYTHON_INLINE PyObject *__Pyx_GetItemInt_Fast(PyObject *o, Py_ssize_t i) {
+    if (PyList_CheckExact(o)) {
+        Py_ssize_t n = (likely(i >= 0)) ? i : i + PyList_GET_SIZE(o);
+        if (likely((n >= 0) & (n < PyList_GET_SIZE(o)))) {
+            PyObject *r = PyList_GET_ITEM(o, n);
+            Py_INCREF(r);
+            return r;
+        }
+    }
+    else if (PyTuple_CheckExact(o)) {
+        Py_ssize_t n = (likely(i >= 0)) ? i : i + PyTuple_GET_SIZE(o);
+        if (likely((n >= 0) & (n < PyTuple_GET_SIZE(o)))) {
+            PyObject *r = PyTuple_GET_ITEM(o, n);
+            Py_INCREF(r);
+            return r;
+        }
+    }
+    else if (likely(i >= 0)) {
+        PySequenceMethods *m = Py_TYPE(o)->tp_as_sequence;
+        if (likely(m && m->sq_item)) {
+            return m->sq_item(o, i);
+        }
+    }
+    return __Pyx_GetItemInt_Generic(o, PyInt_FromSsize_t(i));
+}
+
+static PyObject *__Pyx_GetName(PyObject *dict, PyObject *name); /*proto*/
 
 static CYTHON_INLINE int  __Pyx_GetBufferAndValidate(Py_buffer* buf, PyObject* obj,
     __Pyx_TypeInfo* dtype, int flags, int nd, int cast, __Pyx_BufFmt_StackElem* stack);
 static CYTHON_INLINE void __Pyx_SafeReleaseBuffer(Py_buffer* info);
 
+#define __Pyx_BufPtrStrided2d(type, buf, i0, s0, i1, s1) (type)((char*)buf + i0 * s0 + i1 * s1)
 static CYTHON_INLINE void __Pyx_ErrRestore(PyObject *type, PyObject *value, PyObject *tb); /*proto*/
 static CYTHON_INLINE void __Pyx_ErrFetch(PyObject **type, PyObject **value, PyObject **tb); /*proto*/
 
-static CYTHON_INLINE int __Pyx_CheckKeywordStrings(PyObject *kwdict, const char* function_name, int kw_allowed); /*proto*/
-
 #define __Pyx_BufPtrCContig1d(type, buf, i0, s0) ((type)buf + i0)
 static int __Pyx_ArgTypeTest(PyObject *obj, PyTypeObject *type, int none_allowed,
     const char *name, int exact); /*proto*/
 
-#define __Pyx_BufPtrStrided2d(type, buf, i0, s0, i1, s1) (type)((char*)buf + i0 * s0 + i1 * s1)
 #define __Pyx_BufPtrStrided1d(type, buf, i0, s0) (type)((char*)buf + i0 * s0)
+#define __Pyx_BufPtrStrided3d(type, buf, i0, s0, i1, s1, i2, s2) (type)((char*)buf + i0 * s0 + i1 * s1 + i2 * s2)
 static void __Pyx_RaiseBufferFallbackError(void); /*proto*/
 
 static void __Pyx_Raise(PyObject *type, PyObject *value, PyObject *tb, PyObject *cause); /*proto*/
@@ -990,8 +1061,6 @@ static Py_ssize_t __Pyx_minusones[] = {-1, -1, -1, -1, -1, -1, -1, -1};
 
 static PyObject *__Pyx_Import(PyObject *name, PyObject *from_list, long level); /*proto*/
 
-static CYTHON_INLINE npy_int32 __Pyx_PyInt_from_py_npy_int32(PyObject *);
-
 #if CYTHON_CCOMPLEX
   #ifdef __cplusplus
     #define __Pyx_CREAL(z) ((z).real())
@@ -1187,24 +1256,26 @@ static PyTypeObject *__pyx_ptype_7sklearn_4tree_5_tree_Gini = 0;
 static PyTypeObject *__pyx_ptype_7sklearn_4tree_5_tree_Entropy = 0;
 static PyTypeObject *__pyx_ptype_7sklearn_4tree_5_tree_RegressionCriterion = 0;
 static PyTypeObject *__pyx_ptype_7sklearn_4tree_5_tree_MSE = 0;
+static __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_v_7sklearn_4tree_5_tree_INFINITY;
 static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *, int); /*proto*/
-static __Pyx_TypeInfo __Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t = { "int32_t", NULL, sizeof(__pyx_t_5numpy_int32_t), { 0 }, 0, 'I', IS_UNSIGNED(__pyx_t_5numpy_int32_t), 0 };
+static __Pyx_TypeInfo __Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t = { "DTYPE_t", NULL, sizeof(__pyx_t_7sklearn_4tree_5_tree_DTYPE_t), { 0 }, 0, 'R', 0, 0 };
 static __Pyx_TypeInfo __Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t = { "float64_t", NULL, sizeof(__pyx_t_5numpy_float64_t), { 0 }, 0, 'R', 0, 0 };
 static __Pyx_TypeInfo __Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_BOOL_t = { "BOOL_t", NULL, sizeof(__pyx_t_7sklearn_4tree_5_tree_BOOL_t), { 0 }, 0, 'I', IS_UNSIGNED(__pyx_t_7sklearn_4tree_5_tree_BOOL_t), 0 };
-static __Pyx_TypeInfo __Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t = { "DTYPE_t", NULL, sizeof(__pyx_t_7sklearn_4tree_5_tree_DTYPE_t), { 0 }, 0, 'R', 0, 0 };
+static __Pyx_TypeInfo __Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t = { "int32_t", NULL, sizeof(__pyx_t_5numpy_int32_t), { 0 }, 0, 'I', IS_UNSIGNED(__pyx_t_5numpy_int32_t), 0 };
 #define __Pyx_MODULE_NAME "sklearn.tree._tree"
 int __pyx_module_is_main_sklearn__tree___tree = 0;
 
 /* Implementation of 'sklearn.tree._tree' */
-static PyObject *__pyx_builtin_range;
-static PyObject *__pyx_builtin_xrange;
 static PyObject *__pyx_builtin_ValueError;
+static PyObject *__pyx_builtin_range;
 static PyObject *__pyx_builtin_RuntimeError;
 static PyObject *__pyx_pf_7sklearn_4tree_5_tree_9Criterion_init_value(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *__pyx_v_self); /* proto */
-static int __pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion___init__(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self, int __pyx_v_n_classes); /* proto */
-static PyObject *__pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_2init_value(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self); /* proto */
-static int __pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion___init__(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self); /* proto */
-static PyObject *__pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_2init_value(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self); /* proto */
+static int __pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion___init__(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self, int __pyx_v_n_outputs, PyObject *__pyx_v_n_classes); /* proto */
+static PyObject *__pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_2__del__(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self); /* proto */
+static PyObject *__pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_4init_value(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self); /* proto */
+static int __pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion___init__(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self, int __pyx_v_n_outputs); /* proto */
+static PyObject *__pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_2__del__(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self); /* proto */
+static PyObject *__pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_4init_value(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self); /* proto */
 static PyObject *__pyx_pf_7sklearn_4tree_5_tree__random_sample_mask(CYTHON_UNUSED PyObject *__pyx_self, int __pyx_v_n_total_samples, int __pyx_v_n_total_in_bag, PyObject *__pyx_v_random_state); /* proto */
 static PyObject *__pyx_pf_7sklearn_4tree_5_tree_2_apply_tree(CYTHON_UNUSED PyObject *__pyx_self, PyArrayObject *__pyx_v_X, PyArrayObject *__pyx_v_children, PyArrayObject *__pyx_v_feature, PyArrayObject *__pyx_v_threshold, PyArrayObject *__pyx_v_out); /* proto */
 static PyObject *__pyx_pf_7sklearn_4tree_5_tree_4_predict_tree(CYTHON_UNUSED PyObject *__pyx_self, PyArrayObject *__pyx_v_X, PyArrayObject *__pyx_v_children, PyArrayObject *__pyx_v_feature, PyArrayObject *__pyx_v_threshold, PyArrayObject *__pyx_v_values, PyArrayObject *__pyx_v_pred); /* proto */
@@ -1219,16 +1290,14 @@ static char __pyx_k_5[] = "Non-native byte order not supported";
 static char __pyx_k_7[] = "unknown dtype code in numpy.pxd (%d)";
 static char __pyx_k_8[] = "Format string allocated too short, see comment in numpy.pxd";
 static char __pyx_k_11[] = "Format string allocated too short.";
-static char __pyx_k_15[] = "/home/pprett/workspace/scikit-learn/sklearn/tree/_tree.pyx";
+static char __pyx_k_15[] = "/home/gilles/Sources/scikit-learn/sklearn/tree/_tree.pyx";
 static char __pyx_k_16[] = "sklearn.tree._tree";
 static char __pyx_k_23[] = "X_argsorted_elem_stride";
 static char __pyx_k_24[] = "X_argsorted_col_stride";
 static char __pyx_k_29[] = "_find_best_random_split";
 static char __pyx_k__B[] = "B";
-static char __pyx_k__C[] = "C";
 static char __pyx_k__H[] = "H";
 static char __pyx_k__I[] = "I";
-static char __pyx_k__K[] = "K";
 static char __pyx_k__L[] = "L";
 static char __pyx_k__O[] = "O";
 static char __pyx_k__Q[] = "Q";
@@ -1263,7 +1332,6 @@ static char __pyx_k__dtype[] = "dtype";
 static char __pyx_k__error[] = "error";
 static char __pyx_k__int32[] = "int32";
 static char __pyx_k__numpy[] = "numpy";
-static char __pyx_k__order[] = "order";
 static char __pyx_k__range[] = "range";
 static char __pyx_k__y_ptr[] = "y_ptr";
 static char __pyx_k__zeros[] = "zeros";
@@ -1273,8 +1341,6 @@ static char __pyx_k__best_i[] = "best_i";
 static char __pyx_k__best_t[] = "best_t";
 static char __pyx_k__n_left[] = "n_left";
 static char __pyx_k__values[] = "values";
-static char __pyx_k__xrange[] = "xrange";
-static char __pyx_k__asarray[] = "asarray";
 static char __pyx_k__feature[] = "feature";
 static char __pyx_k__float32[] = "float32";
 static char __pyx_k__node_id[] = "node_id";
@@ -1285,8 +1351,10 @@ static char __pyx_k__children[] = "children";
 static char __pyx_k__features[] = "features";
 static char __pyx_k__min_leaf[] = "min_leaf";
 static char __pyx_k__n_bagged[] = "n_bagged";
+static char __pyx_k__y_stride[] = "y_stride";
 static char __pyx_k__criterion[] = "criterion";
 static char __pyx_k__n_classes[] = "n_classes";
+static char __pyx_k__n_outputs[] = "n_outputs";
 static char __pyx_k__n_samples[] = "n_samples";
 static char __pyx_k__threshold[] = "threshold";
 static char __pyx_k__ValueError[] = "ValueError";
@@ -1324,9 +1392,7 @@ static PyObject *__pyx_kp_u_3;
 static PyObject *__pyx_kp_u_5;
 static PyObject *__pyx_kp_u_7;
 static PyObject *__pyx_kp_u_8;
-static PyObject *__pyx_n_s__C;
 static PyObject *__pyx_n_s__DTYPE;
-static PyObject *__pyx_n_s__K;
 static PyObject *__pyx_n_s__RuntimeError;
 static PyObject *__pyx_n_s__ValueError;
 static PyObject *__pyx_n_s__X;
@@ -1346,7 +1412,6 @@ static PyObject *__pyx_n_s___predict_tree;
 static PyObject *__pyx_n_s___random_sample_mask;
 static PyObject *__pyx_n_s__a;
 static PyObject *__pyx_n_s__arange;
-static PyObject *__pyx_n_s__asarray;
 static PyObject *__pyx_n_s__astype;
 static PyObject *__pyx_n_s__b;
 static PyObject *__pyx_n_s__best_error;
@@ -1376,13 +1441,13 @@ static PyObject *__pyx_n_s__n_bagged;
 static PyObject *__pyx_n_s__n_classes;
 static PyObject *__pyx_n_s__n_features;
 static PyObject *__pyx_n_s__n_left;
+static PyObject *__pyx_n_s__n_outputs;
 static PyObject *__pyx_n_s__n_samples;
 static PyObject *__pyx_n_s__n_total_in_bag;
 static PyObject *__pyx_n_s__n_total_samples;
 static PyObject *__pyx_n_s__node_id;
 static PyObject *__pyx_n_s__np;
 static PyObject *__pyx_n_s__numpy;
-static PyObject *__pyx_n_s__order;
 static PyObject *__pyx_n_s__out;
 static PyObject *__pyx_n_s__permutation;
 static PyObject *__pyx_n_s__pred;
@@ -1394,10 +1459,11 @@ static PyObject *__pyx_n_s__sample_mask_ptr;
 static PyObject *__pyx_n_s__t;
 static PyObject *__pyx_n_s__threshold;
 static PyObject *__pyx_n_s__values;
-static PyObject *__pyx_n_s__xrange;
 static PyObject *__pyx_n_s__y;
 static PyObject *__pyx_n_s__y_ptr;
+static PyObject *__pyx_n_s__y_stride;
 static PyObject *__pyx_n_s__zeros;
+static PyObject *__pyx_int_1;
 static PyObject *__pyx_int_15;
 static PyObject *__pyx_k_tuple_2;
 static PyObject *__pyx_k_tuple_4;
@@ -1418,22 +1484,22 @@ static PyObject *__pyx_k_codeobj_22;
 static PyObject *__pyx_k_codeobj_26;
 static PyObject *__pyx_k_codeobj_28;
 
-/* "sklearn/tree/_tree.pyx":43
+/* "sklearn/tree/_tree.pyx":50
  *     """Interface for splitting criteria (regression and classification)"""
  * 
- *     cdef void init(self, DTYPE_t *y, BOOL_t *sample_mask, int n_samples,             # <<<<<<<<<<<<<<
- *                    int n_total_samples):
- *         """Initialise the criterion class for new split point."""
+ *     cdef void init(self, DTYPE_t* y,             # <<<<<<<<<<<<<<
+ *                          int y_stride,
+ *                          BOOL_t* sample_mask,
  */
 
-static void __pyx_f_7sklearn_4tree_5_tree_9Criterion_init(CYTHON_UNUSED struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *__pyx_v_self, CYTHON_UNUSED __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, CYTHON_UNUSED __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask, CYTHON_UNUSED int __pyx_v_n_samples, CYTHON_UNUSED int __pyx_v_n_total_samples) {
+static void __pyx_f_7sklearn_4tree_5_tree_9Criterion_init(CYTHON_UNUSED struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *__pyx_v_self, CYTHON_UNUSED __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, CYTHON_UNUSED int __pyx_v_y_stride, CYTHON_UNUSED __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask, CYTHON_UNUSED int __pyx_v_n_samples, CYTHON_UNUSED int __pyx_v_n_total_samples) {
   __Pyx_RefNannyDeclarations
   __Pyx_RefNannySetupContext("init", 0);
 
   __Pyx_RefNannyFinishContext();
 }
 
-/* "sklearn/tree/_tree.pyx":48
+/* "sklearn/tree/_tree.pyx":58
  *         pass
  * 
  *     cdef void reset(self):             # <<<<<<<<<<<<<<
@@ -1448,15 +1514,15 @@ static void __pyx_f_7sklearn_4tree_5_tree_9Criterion_reset(CYTHON_UNUSED struct
   __Pyx_RefNannyFinishContext();
 }
 
-/* "sklearn/tree/_tree.pyx":52
+/* "sklearn/tree/_tree.pyx":62
  *         pass
  * 
- *     cdef int update(self, int a, int b, DTYPE_t *y, int *X_argsorted_i,             # <<<<<<<<<<<<<<
- *                     BOOL_t *sample_mask):
- *         """Update the criteria for each value in interval [a,b) (where a and b
+ *     cdef int update(self, int a,             # <<<<<<<<<<<<<<
+ *                           int b,
+ *                           DTYPE_t* y,
  */
 
-static int __pyx_f_7sklearn_4tree_5_tree_9Criterion_update(CYTHON_UNUSED struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *__pyx_v_self, CYTHON_UNUSED int __pyx_v_a, CYTHON_UNUSED int __pyx_v_b, CYTHON_UNUSED __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, CYTHON_UNUSED int *__pyx_v_X_argsorted_i, CYTHON_UNUSED __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask) {
+static int __pyx_f_7sklearn_4tree_5_tree_9Criterion_update(CYTHON_UNUSED struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *__pyx_v_self, CYTHON_UNUSED int __pyx_v_a, CYTHON_UNUSED int __pyx_v_b, CYTHON_UNUSED __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, CYTHON_UNUSED int __pyx_v_y_stride, CYTHON_UNUSED int *__pyx_v_X_argsorted_i, CYTHON_UNUSED __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask) {
   int __pyx_r;
   __Pyx_RefNannyDeclarations
   __Pyx_RefNannySetupContext("update", 0);
@@ -1466,7 +1532,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_9Criterion_update(CYTHON_UNUSED struct
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":58
+/* "sklearn/tree/_tree.pyx":72
  *         pass
  * 
  *     cdef double eval(self):             # <<<<<<<<<<<<<<
@@ -1484,7 +1550,7 @@ static double __pyx_f_7sklearn_4tree_5_tree_9Criterion_eval(CYTHON_UNUSED struct
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":62
+/* "sklearn/tree/_tree.pyx":76
  *         pass
  * 
  *     cpdef np.ndarray init_value(self):             # <<<<<<<<<<<<<<
@@ -1506,13 +1572,13 @@ static PyArrayObject *__pyx_f_7sklearn_4tree_5_tree_9Criterion_init_value(CYTHON
   if (unlikely(__pyx_skip_dispatch)) ;
   /* Check if overriden in Python */
   else if (unlikely(Py_TYPE(((PyObject *)__pyx_v_self))->tp_dictoffset != 0)) {
-    __pyx_t_1 = PyObject_GetAttr(((PyObject *)__pyx_v_self), __pyx_n_s__init_value); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 62; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_1 = PyObject_GetAttr(((PyObject *)__pyx_v_self), __pyx_n_s__init_value); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 76; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     __Pyx_GOTREF(__pyx_t_1);
     if (!PyCFunction_Check(__pyx_t_1) || (PyCFunction_GET_FUNCTION(__pyx_t_1) != (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_9Criterion_1init_value)) {
       __Pyx_XDECREF(((PyObject *)__pyx_r));
-      __pyx_t_2 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_empty_tuple), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 62; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      __pyx_t_2 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_empty_tuple), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 76; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
       __Pyx_GOTREF(__pyx_t_2);
-      if (!(likely(((__pyx_t_2) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_2, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 62; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      if (!(likely(((__pyx_t_2) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_2, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 76; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
       __pyx_r = ((PyArrayObject *)__pyx_t_2);
       __pyx_t_2 = 0;
       __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
@@ -1521,7 +1587,7 @@ static PyArrayObject *__pyx_f_7sklearn_4tree_5_tree_9Criterion_init_value(CYTHON
     __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
   }
 
-  /* "sklearn/tree/_tree.pyx":64
+  /* "sklearn/tree/_tree.pyx":78
  *     cpdef np.ndarray init_value(self):
  *         """Get the init value of the criterion - `init` must be called before."""
  *         pass             # <<<<<<<<<<<<<<
@@ -1554,7 +1620,7 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_9Criterion_1init_value(PyObject
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":62
+/* "sklearn/tree/_tree.pyx":76
  *         pass
  * 
  *     cpdef np.ndarray init_value(self):             # <<<<<<<<<<<<<<
@@ -1571,7 +1637,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_9Criterion_init_value(struct __p
   int __pyx_clineno = 0;
   __Pyx_RefNannySetupContext("init_value", 0);
   __Pyx_XDECREF(__pyx_r);
-  __pyx_t_1 = ((PyObject *)((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_self->__pyx_vtab)->init_value(__pyx_v_self, 1)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 62; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = ((PyObject *)((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_self->__pyx_vtab)->init_value(__pyx_v_self, 1)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 76; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
   __pyx_r = __pyx_t_1;
   __pyx_t_1 = 0;
@@ -1592,17 +1658,19 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_9Criterion_init_value(struct __p
 /* Python wrapper */
 static int __pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_1__init__(PyObject *__pyx_v_self, PyObject *__pyx_args, PyObject *__pyx_kwds); /*proto*/
 static int __pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_1__init__(PyObject *__pyx_v_self, PyObject *__pyx_args, PyObject *__pyx_kwds) {
-  int __pyx_v_n_classes;
-  static PyObject **__pyx_pyargnames[] = {&__pyx_n_s__n_classes,0};
+  int __pyx_v_n_outputs;
+  PyObject *__pyx_v_n_classes = 0;
+  static PyObject **__pyx_pyargnames[] = {&__pyx_n_s__n_outputs,&__pyx_n_s__n_classes,0};
   int __pyx_r;
   __Pyx_RefNannyDeclarations
   __Pyx_RefNannySetupContext("__init__ (wrapper)", 0);
   {
-    PyObject* values[1] = {0};
+    PyObject* values[2] = {0,0};
     if (unlikely(__pyx_kwds)) {
       Py_ssize_t kw_args;
       const Py_ssize_t pos_args = PyTuple_GET_SIZE(__pyx_args);
       switch (pos_args) {
+        case  2: values[1] = PyTuple_GET_ITEM(__pyx_args, 1);
         case  1: values[0] = PyTuple_GET_ITEM(__pyx_args, 0);
         case  0: break;
         default: goto __pyx_L5_argtuple_error;
@@ -1610,396 +1678,381 @@ static int __pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_1__init__(Py
       kw_args = PyDict_Size(__pyx_kwds);
       switch (pos_args) {
         case  0:
-        values[0] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__n_classes);
+        values[0] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__n_outputs);
         if (likely(values[0])) kw_args--;
         else goto __pyx_L5_argtuple_error;
+        case  1:
+        values[1] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__n_classes);
+        if (likely(values[1])) kw_args--;
+        else {
+          __Pyx_RaiseArgtupleInvalid("__init__", 1, 2, 2, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 120; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+        }
       }
       if (unlikely(kw_args > 0)) {
-        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "__init__") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 107; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "__init__") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 120; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
       }
-    } else if (PyTuple_GET_SIZE(__pyx_args) != 1) {
+    } else if (PyTuple_GET_SIZE(__pyx_args) != 2) {
       goto __pyx_L5_argtuple_error;
     } else {
       values[0] = PyTuple_GET_ITEM(__pyx_args, 0);
+      values[1] = PyTuple_GET_ITEM(__pyx_args, 1);
     }
-    __pyx_v_n_classes = __Pyx_PyInt_AsInt(values[0]); if (unlikely((__pyx_v_n_classes == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 107; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_n_outputs = __Pyx_PyInt_AsInt(values[0]); if (unlikely((__pyx_v_n_outputs == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 120; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_n_classes = values[1];
   }
   goto __pyx_L4_argument_unpacking_done;
   __pyx_L5_argtuple_error:;
-  __Pyx_RaiseArgtupleInvalid("__init__", 1, 1, 1, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 107; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+  __Pyx_RaiseArgtupleInvalid("__init__", 1, 2, 2, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 120; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
   __pyx_L3_error:;
   __Pyx_AddTraceback("sklearn.tree._tree.ClassificationCriterion.__init__", __pyx_clineno, __pyx_lineno, __pyx_filename);
   __Pyx_RefNannyFinishContext();
   return -1;
   __pyx_L4_argument_unpacking_done:;
-  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion___init__(((struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *)__pyx_v_self), __pyx_v_n_classes);
+  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion___init__(((struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *)__pyx_v_self), __pyx_v_n_outputs, __pyx_v_n_classes);
   __Pyx_RefNannyFinishContext();
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":107
- *     cdef ndarray_label_count_init
+/* "sklearn/tree/_tree.pyx":120
+ *     cdef int n_right
+ * 
+ *     def __init__(self, int n_outputs, object n_classes):             # <<<<<<<<<<<<<<
+ *         cdef int k = 0
  * 
- *     def __init__(self, int n_classes):             # <<<<<<<<<<<<<<
- *         cdef np.ndarray[np.int32_t, ndim=1] ndarray_label_count_left \
- *             = np.zeros((n_classes,), dtype=np.int32, order='C')
  */
 
-static int __pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion___init__(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self, int __pyx_v_n_classes) {
-  PyArrayObject *__pyx_v_ndarray_label_count_left = 0;
-  PyArrayObject *__pyx_v_ndarray_label_count_right = 0;
-  PyArrayObject *__pyx_v_ndarray_label_count_init = 0;
-  __Pyx_LocalBuf_ND __pyx_pybuffernd_ndarray_label_count_init;
-  __Pyx_Buffer __pyx_pybuffer_ndarray_label_count_init;
-  __Pyx_LocalBuf_ND __pyx_pybuffernd_ndarray_label_count_left;
-  __Pyx_Buffer __pyx_pybuffer_ndarray_label_count_left;
-  __Pyx_LocalBuf_ND __pyx_pybuffernd_ndarray_label_count_right;
-  __Pyx_Buffer __pyx_pybuffer_ndarray_label_count_right;
+static int __pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion___init__(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self, int __pyx_v_n_outputs, PyObject *__pyx_v_n_classes) {
+  int __pyx_v_k;
+  int __pyx_v_label_count_stride;
   int __pyx_r;
   __Pyx_RefNannyDeclarations
-  PyObject *__pyx_t_1 = NULL;
+  int __pyx_t_1;
   PyObject *__pyx_t_2 = NULL;
-  PyObject *__pyx_t_3 = NULL;
+  int __pyx_t_3;
   PyObject *__pyx_t_4 = NULL;
   PyObject *__pyx_t_5 = NULL;
-  PyArrayObject *__pyx_t_6 = NULL;
-  PyArrayObject *__pyx_t_7 = NULL;
-  PyArrayObject *__pyx_t_8 = NULL;
+  int __pyx_t_6;
   int __pyx_lineno = 0;
   const char *__pyx_filename = NULL;
   int __pyx_clineno = 0;
   __Pyx_RefNannySetupContext("__init__", 0);
-  __pyx_pybuffer_ndarray_label_count_left.pybuffer.buf = NULL;
-  __pyx_pybuffer_ndarray_label_count_left.refcount = 0;
-  __pyx_pybuffernd_ndarray_label_count_left.data = NULL;
-  __pyx_pybuffernd_ndarray_label_count_left.rcbuffer = &__pyx_pybuffer_ndarray_label_count_left;
-  __pyx_pybuffer_ndarray_label_count_right.pybuffer.buf = NULL;
-  __pyx_pybuffer_ndarray_label_count_right.refcount = 0;
-  __pyx_pybuffernd_ndarray_label_count_right.data = NULL;
-  __pyx_pybuffernd_ndarray_label_count_right.rcbuffer = &__pyx_pybuffer_ndarray_label_count_right;
-  __pyx_pybuffer_ndarray_label_count_init.pybuffer.buf = NULL;
-  __pyx_pybuffer_ndarray_label_count_init.refcount = 0;
-  __pyx_pybuffernd_ndarray_label_count_init.data = NULL;
-  __pyx_pybuffernd_ndarray_label_count_init.rcbuffer = &__pyx_pybuffer_ndarray_label_count_init;
-
-  /* "sklearn/tree/_tree.pyx":109
- *     def __init__(self, int n_classes):
- *         cdef np.ndarray[np.int32_t, ndim=1] ndarray_label_count_left \
- *             = np.zeros((n_classes,), dtype=np.int32, order='C')             # <<<<<<<<<<<<<<
- *         cdef np.ndarray[np.int32_t, ndim=1] ndarray_label_count_right \
- *             = np.zeros((n_classes,), dtype=np.int32, order='C')
- */
-  __pyx_t_1 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_2 = PyObject_GetAttr(__pyx_t_1, __pyx_n_s__zeros); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __pyx_t_1 = PyInt_FromLong(__pyx_v_n_classes); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_3 = PyTuple_New(1); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_3);
-  PyTuple_SET_ITEM(__pyx_t_3, 0, __pyx_t_1);
-  __Pyx_GIVEREF(__pyx_t_1);
-  __pyx_t_1 = 0;
-  __pyx_t_1 = PyTuple_New(1); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  PyTuple_SET_ITEM(__pyx_t_1, 0, ((PyObject *)__pyx_t_3));
-  __Pyx_GIVEREF(((PyObject *)__pyx_t_3));
-  __pyx_t_3 = 0;
-  __pyx_t_3 = PyDict_New(); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(((PyObject *)__pyx_t_3));
-  __pyx_t_4 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_4);
-  __pyx_t_5 = PyObject_GetAttr(__pyx_t_4, __pyx_n_s__int32); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_5);
-  __Pyx_DECREF(__pyx_t_4); __pyx_t_4 = 0;
-  if (PyDict_SetItem(__pyx_t_3, ((PyObject *)__pyx_n_s__dtype), __pyx_t_5) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
-  if (PyDict_SetItem(__pyx_t_3, ((PyObject *)__pyx_n_s__order), ((PyObject *)__pyx_n_s__C)) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __pyx_t_5 = PyObject_Call(__pyx_t_2, ((PyObject *)__pyx_t_1), ((PyObject *)__pyx_t_3)); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_5);
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_1)); __pyx_t_1 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_3)); __pyx_t_3 = 0;
-  if (!(likely(((__pyx_t_5) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_5, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 109; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __pyx_t_6 = ((PyArrayObject *)__pyx_t_5);
-  {
-    __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_ndarray_label_count_left.rcbuffer->pybuffer, (PyObject*)__pyx_t_6, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {
-      __pyx_v_ndarray_label_count_left = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None); __pyx_pybuffernd_ndarray_label_count_left.rcbuffer->pybuffer.buf = NULL;
-      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 108; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    } else {__pyx_pybuffernd_ndarray_label_count_left.diminfo[0].strides = __pyx_pybuffernd_ndarray_label_count_left.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_ndarray_label_count_left.diminfo[0].shape = __pyx_pybuffernd_ndarray_label_count_left.rcbuffer->pybuffer.shape[0];
-    }
-  }
-  __pyx_t_6 = 0;
-  __pyx_v_ndarray_label_count_left = ((PyArrayObject *)__pyx_t_5);
-  __pyx_t_5 = 0;
 
-  /* "sklearn/tree/_tree.pyx":111
- *             = np.zeros((n_classes,), dtype=np.int32, order='C')
- *         cdef np.ndarray[np.int32_t, ndim=1] ndarray_label_count_right \
- *             = np.zeros((n_classes,), dtype=np.int32, order='C')             # <<<<<<<<<<<<<<
- *         cdef np.ndarray[np.int32_t, ndim=1] ndarray_label_count_init \
- *             = np.zeros((n_classes,), dtype=np.int32, order='C')
+  /* "sklearn/tree/_tree.pyx":121
+ * 
+ *     def __init__(self, int n_outputs, object n_classes):
+ *         cdef int k = 0             # <<<<<<<<<<<<<<
+ * 
+ *         self.n_outputs = n_outputs
  */
-  __pyx_t_5 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_5);
-  __pyx_t_3 = PyObject_GetAttr(__pyx_t_5, __pyx_n_s__zeros); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_3);
-  __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
-  __pyx_t_5 = PyInt_FromLong(__pyx_v_n_classes); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_5);
-  __pyx_t_1 = PyTuple_New(1); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  PyTuple_SET_ITEM(__pyx_t_1, 0, __pyx_t_5);
-  __Pyx_GIVEREF(__pyx_t_5);
-  __pyx_t_5 = 0;
-  __pyx_t_5 = PyTuple_New(1); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_5);
-  PyTuple_SET_ITEM(__pyx_t_5, 0, ((PyObject *)__pyx_t_1));
-  __Pyx_GIVEREF(((PyObject *)__pyx_t_1));
-  __pyx_t_1 = 0;
-  __pyx_t_1 = PyDict_New(); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(((PyObject *)__pyx_t_1));
-  __pyx_t_2 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __pyx_t_4 = PyObject_GetAttr(__pyx_t_2, __pyx_n_s__int32); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_4);
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  if (PyDict_SetItem(__pyx_t_1, ((PyObject *)__pyx_n_s__dtype), __pyx_t_4) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_4); __pyx_t_4 = 0;
-  if (PyDict_SetItem(__pyx_t_1, ((PyObject *)__pyx_n_s__order), ((PyObject *)__pyx_n_s__C)) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __pyx_t_4 = PyObject_Call(__pyx_t_3, ((PyObject *)__pyx_t_5), ((PyObject *)__pyx_t_1)); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_4);
-  __Pyx_DECREF(__pyx_t_3); __pyx_t_3 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_5)); __pyx_t_5 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_1)); __pyx_t_1 = 0;
-  if (!(likely(((__pyx_t_4) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_4, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 111; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __pyx_t_7 = ((PyArrayObject *)__pyx_t_4);
-  {
-    __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_ndarray_label_count_right.rcbuffer->pybuffer, (PyObject*)__pyx_t_7, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {
-      __pyx_v_ndarray_label_count_right = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None); __pyx_pybuffernd_ndarray_label_count_right.rcbuffer->pybuffer.buf = NULL;
-      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 110; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    } else {__pyx_pybuffernd_ndarray_label_count_right.diminfo[0].strides = __pyx_pybuffernd_ndarray_label_count_right.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_ndarray_label_count_right.diminfo[0].shape = __pyx_pybuffernd_ndarray_label_count_right.rcbuffer->pybuffer.shape[0];
-    }
-  }
-  __pyx_t_7 = 0;
-  __pyx_v_ndarray_label_count_right = ((PyArrayObject *)__pyx_t_4);
-  __pyx_t_4 = 0;
+  __pyx_v_k = 0;
 
-  /* "sklearn/tree/_tree.pyx":113
- *             = np.zeros((n_classes,), dtype=np.int32, order='C')
- *         cdef np.ndarray[np.int32_t, ndim=1] ndarray_label_count_init \
- *             = np.zeros((n_classes,), dtype=np.int32, order='C')             # <<<<<<<<<<<<<<
+  /* "sklearn/tree/_tree.pyx":123
+ *         cdef int k = 0
  * 
- *         self.n_classes = n_classes
+ *         self.n_outputs = n_outputs             # <<<<<<<<<<<<<<
+ *         self.n_classes = <int*> calloc(n_outputs, sizeof(int))
+ *         cdef int label_count_stride = -1
  */
-  __pyx_t_4 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_4);
-  __pyx_t_1 = PyObject_GetAttr(__pyx_t_4, __pyx_n_s__zeros); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __Pyx_DECREF(__pyx_t_4); __pyx_t_4 = 0;
-  __pyx_t_4 = PyInt_FromLong(__pyx_v_n_classes); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_4);
-  __pyx_t_5 = PyTuple_New(1); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_5);
-  PyTuple_SET_ITEM(__pyx_t_5, 0, __pyx_t_4);
-  __Pyx_GIVEREF(__pyx_t_4);
-  __pyx_t_4 = 0;
-  __pyx_t_4 = PyTuple_New(1); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_4);
-  PyTuple_SET_ITEM(__pyx_t_4, 0, ((PyObject *)__pyx_t_5));
-  __Pyx_GIVEREF(((PyObject *)__pyx_t_5));
-  __pyx_t_5 = 0;
-  __pyx_t_5 = PyDict_New(); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(((PyObject *)__pyx_t_5));
-  __pyx_t_3 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_3);
-  __pyx_t_2 = PyObject_GetAttr(__pyx_t_3, __pyx_n_s__int32); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __Pyx_DECREF(__pyx_t_3); __pyx_t_3 = 0;
-  if (PyDict_SetItem(__pyx_t_5, ((PyObject *)__pyx_n_s__dtype), __pyx_t_2) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  if (PyDict_SetItem(__pyx_t_5, ((PyObject *)__pyx_n_s__order), ((PyObject *)__pyx_n_s__C)) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __pyx_t_2 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_t_4), ((PyObject *)__pyx_t_5)); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_4)); __pyx_t_4 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_5)); __pyx_t_5 = 0;
-  if (!(likely(((__pyx_t_2) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_2, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 113; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __pyx_t_8 = ((PyArrayObject *)__pyx_t_2);
-  {
-    __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_ndarray_label_count_init.rcbuffer->pybuffer, (PyObject*)__pyx_t_8, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {
-      __pyx_v_ndarray_label_count_init = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None); __pyx_pybuffernd_ndarray_label_count_init.rcbuffer->pybuffer.buf = NULL;
-      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 112; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    } else {__pyx_pybuffernd_ndarray_label_count_init.diminfo[0].strides = __pyx_pybuffernd_ndarray_label_count_init.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_ndarray_label_count_init.diminfo[0].shape = __pyx_pybuffernd_ndarray_label_count_init.rcbuffer->pybuffer.shape[0];
+  __pyx_v_self->n_outputs = __pyx_v_n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":124
+ * 
+ *         self.n_outputs = n_outputs
+ *         self.n_classes = <int*> calloc(n_outputs, sizeof(int))             # <<<<<<<<<<<<<<
+ *         cdef int label_count_stride = -1
+ * 
+ */
+  __pyx_v_self->n_classes = ((int *)calloc(__pyx_v_n_outputs, (sizeof(int))));
+
+  /* "sklearn/tree/_tree.pyx":125
+ *         self.n_outputs = n_outputs
+ *         self.n_classes = <int*> calloc(n_outputs, sizeof(int))
+ *         cdef int label_count_stride = -1             # <<<<<<<<<<<<<<
+ * 
+ *         for k from 0 <= k < n_outputs:
+ */
+  __pyx_v_label_count_stride = -1;
+
+  /* "sklearn/tree/_tree.pyx":127
+ *         cdef int label_count_stride = -1
+ * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             self.n_classes[k] = n_classes[k]
+ * 
+ */
+  __pyx_t_1 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
+
+    /* "sklearn/tree/_tree.pyx":128
+ * 
+ *         for k from 0 <= k < n_outputs:
+ *             self.n_classes[k] = n_classes[k]             # <<<<<<<<<<<<<<
+ * 
+ *             if n_classes[k] > label_count_stride:
+ */
+    __pyx_t_2 = __Pyx_GetItemInt(__pyx_v_n_classes, __pyx_v_k, sizeof(int), PyInt_FromLong); if (!__pyx_t_2) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 128; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_2);
+    __pyx_t_3 = __Pyx_PyInt_AsInt(__pyx_t_2); if (unlikely((__pyx_t_3 == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 128; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+    (__pyx_v_self->n_classes[__pyx_v_k]) = __pyx_t_3;
+
+    /* "sklearn/tree/_tree.pyx":130
+ *             self.n_classes[k] = n_classes[k]
+ * 
+ *             if n_classes[k] > label_count_stride:             # <<<<<<<<<<<<<<
+ *                 label_count_stride = n_classes[k]
+ * 
+ */
+    __pyx_t_2 = __Pyx_GetItemInt(__pyx_v_n_classes, __pyx_v_k, sizeof(int), PyInt_FromLong); if (!__pyx_t_2) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 130; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_2);
+    __pyx_t_4 = PyInt_FromLong(__pyx_v_label_count_stride); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 130; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_4);
+    __pyx_t_5 = PyObject_RichCompare(__pyx_t_2, __pyx_t_4, Py_GT); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 130; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_5);
+    __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+    __Pyx_DECREF(__pyx_t_4); __pyx_t_4 = 0;
+    __pyx_t_6 = __Pyx_PyObject_IsTrue(__pyx_t_5); if (unlikely(__pyx_t_6 < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 130; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
+    if (__pyx_t_6) {
+
+      /* "sklearn/tree/_tree.pyx":131
+ * 
+ *             if n_classes[k] > label_count_stride:
+ *                 label_count_stride = n_classes[k]             # <<<<<<<<<<<<<<
+ * 
+ *         self.label_count_stride = label_count_stride
+ */
+      __pyx_t_5 = __Pyx_GetItemInt(__pyx_v_n_classes, __pyx_v_k, sizeof(int), PyInt_FromLong); if (!__pyx_t_5) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 131; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      __Pyx_GOTREF(__pyx_t_5);
+      __pyx_t_3 = __Pyx_PyInt_AsInt(__pyx_t_5); if (unlikely((__pyx_t_3 == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 131; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
+      __pyx_v_label_count_stride = __pyx_t_3;
+      goto __pyx_L5;
     }
+    __pyx_L5:;
   }
-  __pyx_t_8 = 0;
-  __pyx_v_ndarray_label_count_init = ((PyArrayObject *)__pyx_t_2);
-  __pyx_t_2 = 0;
 
-  /* "sklearn/tree/_tree.pyx":115
- *             = np.zeros((n_classes,), dtype=np.int32, order='C')
+  /* "sklearn/tree/_tree.pyx":133
+ *                 label_count_stride = n_classes[k]
+ * 
+ *         self.label_count_stride = label_count_stride             # <<<<<<<<<<<<<<
+ *         self.label_count_left = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+ *         self.label_count_right = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+ */
+  __pyx_v_self->label_count_stride = __pyx_v_label_count_stride;
+
+  /* "sklearn/tree/_tree.pyx":134
+ * 
+ *         self.label_count_stride = label_count_stride
+ *         self.label_count_left = <int*> calloc(n_outputs * label_count_stride, sizeof(int))             # <<<<<<<<<<<<<<
+ *         self.label_count_right = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+ *         self.label_count_init = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+ */
+  __pyx_v_self->label_count_left = ((int *)calloc((__pyx_v_n_outputs * __pyx_v_label_count_stride), (sizeof(int))));
+
+  /* "sklearn/tree/_tree.pyx":135
+ *         self.label_count_stride = label_count_stride
+ *         self.label_count_left = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+ *         self.label_count_right = <int*> calloc(n_outputs * label_count_stride, sizeof(int))             # <<<<<<<<<<<<<<
+ *         self.label_count_init = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+ * 
+ */
+  __pyx_v_self->label_count_right = ((int *)calloc((__pyx_v_n_outputs * __pyx_v_label_count_stride), (sizeof(int))));
+
+  /* "sklearn/tree/_tree.pyx":136
+ *         self.label_count_left = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+ *         self.label_count_right = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+ *         self.label_count_init = <int*> calloc(n_outputs * label_count_stride, sizeof(int))             # <<<<<<<<<<<<<<
  * 
- *         self.n_classes = n_classes             # <<<<<<<<<<<<<<
  *         self.n_samples = 0
- *         self.n_left = 0
  */
-  __pyx_v_self->n_classes = __pyx_v_n_classes;
+  __pyx_v_self->label_count_init = ((int *)calloc((__pyx_v_n_outputs * __pyx_v_label_count_stride), (sizeof(int))));
 
-  /* "sklearn/tree/_tree.pyx":116
+  /* "sklearn/tree/_tree.pyx":138
+ *         self.label_count_init = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
  * 
- *         self.n_classes = n_classes
  *         self.n_samples = 0             # <<<<<<<<<<<<<<
  *         self.n_left = 0
  *         self.n_right = 0
  */
   __pyx_v_self->n_samples = 0;
 
-  /* "sklearn/tree/_tree.pyx":117
- *         self.n_classes = n_classes
+  /* "sklearn/tree/_tree.pyx":139
+ * 
  *         self.n_samples = 0
  *         self.n_left = 0             # <<<<<<<<<<<<<<
  *         self.n_right = 0
- *         self.label_count_left = <int*>ndarray_label_count_left.data
+ * 
  */
   __pyx_v_self->n_left = 0;
 
-  /* "sklearn/tree/_tree.pyx":118
+  /* "sklearn/tree/_tree.pyx":140
  *         self.n_samples = 0
  *         self.n_left = 0
  *         self.n_right = 0             # <<<<<<<<<<<<<<
- *         self.label_count_left = <int*>ndarray_label_count_left.data
- *         self.label_count_right = <int*>ndarray_label_count_right.data
+ * 
+ *     def __del__(self):
  */
   __pyx_v_self->n_right = 0;
 
-  /* "sklearn/tree/_tree.pyx":119
- *         self.n_left = 0
+  __pyx_r = 0;
+  goto __pyx_L0;
+  __pyx_L1_error:;
+  __Pyx_XDECREF(__pyx_t_2);
+  __Pyx_XDECREF(__pyx_t_4);
+  __Pyx_XDECREF(__pyx_t_5);
+  __Pyx_AddTraceback("sklearn.tree._tree.ClassificationCriterion.__init__", __pyx_clineno, __pyx_lineno, __pyx_filename);
+  __pyx_r = -1;
+  __pyx_L0:;
+  __Pyx_RefNannyFinishContext();
+  return __pyx_r;
+}
+
+/* Python wrapper */
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_3__del__(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_3__del__(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused) {
+  PyObject *__pyx_r = 0;
+  __Pyx_RefNannyDeclarations
+  __Pyx_RefNannySetupContext("__del__ (wrapper)", 0);
+  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_2__del__(((struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *)__pyx_v_self));
+  __Pyx_RefNannyFinishContext();
+  return __pyx_r;
+}
+
+/* "sklearn/tree/_tree.pyx":142
  *         self.n_right = 0
- *         self.label_count_left = <int*>ndarray_label_count_left.data             # <<<<<<<<<<<<<<
- *         self.label_count_right = <int*>ndarray_label_count_right.data
- *         self.label_count_init = <int*>ndarray_label_count_init.data
+ * 
+ *     def __del__(self):             # <<<<<<<<<<<<<<
+ *         free(self.n_classes)
+ *         free(self.label_count_left)
  */
-  __pyx_v_self->label_count_left = ((int *)__pyx_v_ndarray_label_count_left->data);
 
-  /* "sklearn/tree/_tree.pyx":120
- *         self.n_right = 0
- *         self.label_count_left = <int*>ndarray_label_count_left.data
- *         self.label_count_right = <int*>ndarray_label_count_right.data             # <<<<<<<<<<<<<<
- *         self.label_count_init = <int*>ndarray_label_count_init.data
- *         self.ndarray_label_count_left = ndarray_label_count_left
+static PyObject *__pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_2__del__(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self) {
+  PyObject *__pyx_r = NULL;
+  __Pyx_RefNannyDeclarations
+  __Pyx_RefNannySetupContext("__del__", 0);
+
+  /* "sklearn/tree/_tree.pyx":143
+ * 
+ *     def __del__(self):
+ *         free(self.n_classes)             # <<<<<<<<<<<<<<
+ *         free(self.label_count_left)
+ *         free(self.label_count_right)
  */
-  __pyx_v_self->label_count_right = ((int *)__pyx_v_ndarray_label_count_right->data);
+  free(__pyx_v_self->n_classes);
 
-  /* "sklearn/tree/_tree.pyx":121
- *         self.label_count_left = <int*>ndarray_label_count_left.data
- *         self.label_count_right = <int*>ndarray_label_count_right.data
- *         self.label_count_init = <int*>ndarray_label_count_init.data             # <<<<<<<<<<<<<<
- *         self.ndarray_label_count_left = ndarray_label_count_left
- *         self.ndarray_label_count_right = ndarray_label_count_right
- */
-  __pyx_v_self->label_count_init = ((int *)__pyx_v_ndarray_label_count_init->data);
-
-  /* "sklearn/tree/_tree.pyx":122
- *         self.label_count_right = <int*>ndarray_label_count_right.data
- *         self.label_count_init = <int*>ndarray_label_count_init.data
- *         self.ndarray_label_count_left = ndarray_label_count_left             # <<<<<<<<<<<<<<
- *         self.ndarray_label_count_right = ndarray_label_count_right
- *         self.ndarray_label_count_init = ndarray_label_count_init
- */
-  __Pyx_INCREF(((PyObject *)__pyx_v_ndarray_label_count_left));
-  __Pyx_GIVEREF(((PyObject *)__pyx_v_ndarray_label_count_left));
-  __Pyx_GOTREF(__pyx_v_self->ndarray_label_count_left);
-  __Pyx_DECREF(__pyx_v_self->ndarray_label_count_left);
-  __pyx_v_self->ndarray_label_count_left = ((PyObject *)__pyx_v_ndarray_label_count_left);
+  /* "sklearn/tree/_tree.pyx":144
+ *     def __del__(self):
+ *         free(self.n_classes)
+ *         free(self.label_count_left)             # <<<<<<<<<<<<<<
+ *         free(self.label_count_right)
+ *         free(self.label_count_init)
+ */
+  free(__pyx_v_self->label_count_left);
 
-  /* "sklearn/tree/_tree.pyx":123
- *         self.label_count_init = <int*>ndarray_label_count_init.data
- *         self.ndarray_label_count_left = ndarray_label_count_left
- *         self.ndarray_label_count_right = ndarray_label_count_right             # <<<<<<<<<<<<<<
- *         self.ndarray_label_count_init = ndarray_label_count_init
+  /* "sklearn/tree/_tree.pyx":145
+ *         free(self.n_classes)
+ *         free(self.label_count_left)
+ *         free(self.label_count_right)             # <<<<<<<<<<<<<<
+ *         free(self.label_count_init)
  * 
  */
-  __Pyx_INCREF(((PyObject *)__pyx_v_ndarray_label_count_right));
-  __Pyx_GIVEREF(((PyObject *)__pyx_v_ndarray_label_count_right));
-  __Pyx_GOTREF(__pyx_v_self->ndarray_label_count_right);
-  __Pyx_DECREF(__pyx_v_self->ndarray_label_count_right);
-  __pyx_v_self->ndarray_label_count_right = ((PyObject *)__pyx_v_ndarray_label_count_right);
+  free(__pyx_v_self->label_count_right);
 
-  /* "sklearn/tree/_tree.pyx":124
- *         self.ndarray_label_count_left = ndarray_label_count_left
- *         self.ndarray_label_count_right = ndarray_label_count_right
- *         self.ndarray_label_count_init = ndarray_label_count_init             # <<<<<<<<<<<<<<
+  /* "sklearn/tree/_tree.pyx":146
+ *         free(self.label_count_left)
+ *         free(self.label_count_right)
+ *         free(self.label_count_init)             # <<<<<<<<<<<<<<
  * 
- *     cdef void init(self, DTYPE_t *y, BOOL_t *sample_mask, int n_samples,
+ *     cdef void init(self, DTYPE_t* y,
  */
-  __Pyx_INCREF(((PyObject *)__pyx_v_ndarray_label_count_init));
-  __Pyx_GIVEREF(((PyObject *)__pyx_v_ndarray_label_count_init));
-  __Pyx_GOTREF(__pyx_v_self->ndarray_label_count_init);
-  __Pyx_DECREF(__pyx_v_self->ndarray_label_count_init);
-  __pyx_v_self->ndarray_label_count_init = ((PyObject *)__pyx_v_ndarray_label_count_init);
+  free(__pyx_v_self->label_count_init);
 
-  __pyx_r = 0;
-  goto __pyx_L0;
-  __pyx_L1_error:;
-  __Pyx_XDECREF(__pyx_t_1);
-  __Pyx_XDECREF(__pyx_t_2);
-  __Pyx_XDECREF(__pyx_t_3);
-  __Pyx_XDECREF(__pyx_t_4);
-  __Pyx_XDECREF(__pyx_t_5);
-  { PyObject *__pyx_type, *__pyx_value, *__pyx_tb;
-    __Pyx_ErrFetch(&__pyx_type, &__pyx_value, &__pyx_tb);
-    __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_ndarray_label_count_init.rcbuffer->pybuffer);
-    __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_ndarray_label_count_left.rcbuffer->pybuffer);
-    __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_ndarray_label_count_right.rcbuffer->pybuffer);
-  __Pyx_ErrRestore(__pyx_type, __pyx_value, __pyx_tb);}
-  __Pyx_AddTraceback("sklearn.tree._tree.ClassificationCriterion.__init__", __pyx_clineno, __pyx_lineno, __pyx_filename);
-  __pyx_r = -1;
-  goto __pyx_L2;
-  __pyx_L0:;
-  __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_ndarray_label_count_init.rcbuffer->pybuffer);
-  __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_ndarray_label_count_left.rcbuffer->pybuffer);
-  __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_ndarray_label_count_right.rcbuffer->pybuffer);
-  __pyx_L2:;
-  __Pyx_XDECREF((PyObject *)__pyx_v_ndarray_label_count_left);
-  __Pyx_XDECREF((PyObject *)__pyx_v_ndarray_label_count_right);
-  __Pyx_XDECREF((PyObject *)__pyx_v_ndarray_label_count_init);
+  __pyx_r = Py_None; __Pyx_INCREF(Py_None);
+  __Pyx_XGIVEREF(__pyx_r);
   __Pyx_RefNannyFinishContext();
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":126
- *         self.ndarray_label_count_init = ndarray_label_count_init
+/* "sklearn/tree/_tree.pyx":148
+ *         free(self.label_count_init)
  * 
- *     cdef void init(self, DTYPE_t *y, BOOL_t *sample_mask, int n_samples,             # <<<<<<<<<<<<<<
- *                    int n_total_samples):
- *         """Initialise the criterion class."""
+ *     cdef void init(self, DTYPE_t* y,             # <<<<<<<<<<<<<<
+ *                          int y_stride,
+ *                          BOOL_t *sample_mask,
  */
 
-static void __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask, int __pyx_v_n_samples, int __pyx_v_n_total_samples) {
+static void __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, int __pyx_v_y_stride, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask, int __pyx_v_n_samples, int __pyx_v_n_total_samples) {
+  int __pyx_v_n_outputs;
+  int *__pyx_v_n_classes;
+  int __pyx_v_label_count_stride;
+  int *__pyx_v_label_count_init;
+  int __pyx_v_k;
   int __pyx_v_c;
   int __pyx_v_j;
   __Pyx_RefNannyDeclarations
   int __pyx_t_1;
   int __pyx_t_2;
   int __pyx_t_3;
+  int __pyx_t_4;
   __Pyx_RefNannySetupContext("init", 0);
 
-  /* "sklearn/tree/_tree.pyx":129
- *                    int n_total_samples):
+  /* "sklearn/tree/_tree.pyx":154
+ *                          int n_total_samples):
+ *         """Initialise the criterion class."""
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
+ */
+  __pyx_v_n_outputs = __pyx_v_self->n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":155
  *         """Initialise the criterion class."""
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes             # <<<<<<<<<<<<<<
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_init = self.label_count_init
+ */
+  __pyx_v_n_classes = __pyx_v_self->n_classes;
+
+  /* "sklearn/tree/_tree.pyx":156
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_init = self.label_count_init
+ * 
+ */
+  __pyx_v_label_count_stride = __pyx_v_self->label_count_stride;
+
+  /* "sklearn/tree/_tree.pyx":157
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_init = self.label_count_init             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int k = 0
+ */
+  __pyx_v_label_count_init = __pyx_v_self->label_count_init;
+
+  /* "sklearn/tree/_tree.pyx":159
+ *         cdef int* label_count_init = self.label_count_init
+ * 
+ *         cdef int k = 0             # <<<<<<<<<<<<<<
+ *         cdef int c = 0
+ *         cdef int j = 0
+ */
+  __pyx_v_k = 0;
+
+  /* "sklearn/tree/_tree.pyx":160
+ * 
+ *         cdef int k = 0
  *         cdef int c = 0             # <<<<<<<<<<<<<<
  *         cdef int j = 0
  * 
  */
   __pyx_v_c = 0;
 
-  /* "sklearn/tree/_tree.pyx":130
- *         """Initialise the criterion class."""
+  /* "sklearn/tree/_tree.pyx":161
+ *         cdef int k = 0
  *         cdef int c = 0
  *         cdef int j = 0             # <<<<<<<<<<<<<<
  * 
@@ -2007,37 +2060,48 @@ static void __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init(struct
  */
   __pyx_v_j = 0;
 
-  /* "sklearn/tree/_tree.pyx":132
+  /* "sklearn/tree/_tree.pyx":163
  *         cdef int j = 0
  * 
  *         self.n_samples = n_samples             # <<<<<<<<<<<<<<
  * 
- *         for c from 0 <= c < self.n_classes:
+ *         for k from 0 <= k < n_outputs:
  */
   __pyx_v_self->n_samples = __pyx_v_n_samples;
 
-  /* "sklearn/tree/_tree.pyx":134
+  /* "sklearn/tree/_tree.pyx":165
  *         self.n_samples = n_samples
  * 
- *         for c from 0 <= c < self.n_classes:             # <<<<<<<<<<<<<<
- *             self.label_count_init[c] = 0
- * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             for c from 0 <= c < n_classes[k]:
+ *                 label_count_init[k * label_count_stride + c] = 0
  */
-  __pyx_t_1 = __pyx_v_self->n_classes;
-  for (__pyx_v_c = 0; __pyx_v_c < __pyx_t_1; __pyx_v_c++) {
+  __pyx_t_1 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
 
-    /* "sklearn/tree/_tree.pyx":135
+    /* "sklearn/tree/_tree.pyx":166
  * 
- *         for c from 0 <= c < self.n_classes:
- *             self.label_count_init[c] = 0             # <<<<<<<<<<<<<<
+ *         for k from 0 <= k < n_outputs:
+ *             for c from 0 <= c < n_classes[k]:             # <<<<<<<<<<<<<<
+ *                 label_count_init[k * label_count_stride + c] = 0
+ * 
+ */
+    __pyx_t_2 = (__pyx_v_n_classes[__pyx_v_k]);
+    for (__pyx_v_c = 0; __pyx_v_c < __pyx_t_2; __pyx_v_c++) {
+
+      /* "sklearn/tree/_tree.pyx":167
+ *         for k from 0 <= k < n_outputs:
+ *             for c from 0 <= c < n_classes[k]:
+ *                 label_count_init[k * label_count_stride + c] = 0             # <<<<<<<<<<<<<<
  * 
  *         for j from 0 <= j < n_total_samples:
  */
-    (__pyx_v_self->label_count_init[__pyx_v_c]) = 0;
+      (__pyx_v_label_count_init[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]) = 0;
+    }
   }
 
-  /* "sklearn/tree/_tree.pyx":137
- *             self.label_count_init[c] = 0
+  /* "sklearn/tree/_tree.pyx":169
+ *                 label_count_init[k * label_count_stride + c] = 0
  * 
  *         for j from 0 <= j < n_total_samples:             # <<<<<<<<<<<<<<
  *             if sample_mask[j] == 0:
@@ -2046,51 +2110,62 @@ static void __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init(struct
   __pyx_t_1 = __pyx_v_n_total_samples;
   for (__pyx_v_j = 0; __pyx_v_j < __pyx_t_1; __pyx_v_j++) {
 
-    /* "sklearn/tree/_tree.pyx":138
+    /* "sklearn/tree/_tree.pyx":170
  * 
  *         for j from 0 <= j < n_total_samples:
  *             if sample_mask[j] == 0:             # <<<<<<<<<<<<<<
  *                 continue
- *             c = <int>(y[j])
+ * 
  */
-    __pyx_t_2 = ((__pyx_v_sample_mask[__pyx_v_j]) == 0);
-    if (__pyx_t_2) {
+    __pyx_t_3 = ((__pyx_v_sample_mask[__pyx_v_j]) == 0);
+    if (__pyx_t_3) {
 
-      /* "sklearn/tree/_tree.pyx":139
+      /* "sklearn/tree/_tree.pyx":171
  *         for j from 0 <= j < n_total_samples:
  *             if sample_mask[j] == 0:
  *                 continue             # <<<<<<<<<<<<<<
- *             c = <int>(y[j])
- *             self.label_count_init[c] += 1
+ * 
+ *             for k from 0 <= k < n_outputs:
  */
-      goto __pyx_L5_continue;
-      goto __pyx_L7;
+      goto __pyx_L7_continue;
+      goto __pyx_L9;
     }
-    __pyx_L7:;
+    __pyx_L9:;
 
-    /* "sklearn/tree/_tree.pyx":140
- *             if sample_mask[j] == 0:
+    /* "sklearn/tree/_tree.pyx":173
  *                 continue
- *             c = <int>(y[j])             # <<<<<<<<<<<<<<
- *             self.label_count_init[c] += 1
  * 
+ *             for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *                 c = <int>y[j * y_stride + k]
+ *                 label_count_init[k * label_count_stride + c] += 1
  */
-    __pyx_v_c = ((int)(__pyx_v_y[__pyx_v_j]));
+    __pyx_t_2 = __pyx_v_n_outputs;
+    for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_2; __pyx_v_k++) {
 
-    /* "sklearn/tree/_tree.pyx":141
- *                 continue
- *             c = <int>(y[j])
- *             self.label_count_init[c] += 1             # <<<<<<<<<<<<<<
+      /* "sklearn/tree/_tree.pyx":174
+ * 
+ *             for k from 0 <= k < n_outputs:
+ *                 c = <int>y[j * y_stride + k]             # <<<<<<<<<<<<<<
+ *                 label_count_init[k * label_count_stride + c] += 1
+ * 
+ */
+      __pyx_v_c = ((int)(__pyx_v_y[((__pyx_v_j * __pyx_v_y_stride) + __pyx_v_k)]));
+
+      /* "sklearn/tree/_tree.pyx":175
+ *             for k from 0 <= k < n_outputs:
+ *                 c = <int>y[j * y_stride + k]
+ *                 label_count_init[k * label_count_stride + c] += 1             # <<<<<<<<<<<<<<
  * 
  *         self.reset()
  */
-    __pyx_t_3 = __pyx_v_c;
-    (__pyx_v_self->label_count_init[__pyx_t_3]) = ((__pyx_v_self->label_count_init[__pyx_t_3]) + 1);
-    __pyx_L5_continue:;
+      __pyx_t_4 = ((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c);
+      (__pyx_v_label_count_init[__pyx_t_4]) = ((__pyx_v_label_count_init[__pyx_t_4]) + 1);
+    }
+    __pyx_L7_continue:;
   }
 
-  /* "sklearn/tree/_tree.pyx":143
- *             self.label_count_init[c] += 1
+  /* "sklearn/tree/_tree.pyx":177
+ *                 label_count_init[k * label_count_stride + c] += 1
  * 
  *         self.reset()             # <<<<<<<<<<<<<<
  * 
@@ -2101,7 +2176,7 @@ static void __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init(struct
   __Pyx_RefNannyFinishContext();
 }
 
-/* "sklearn/tree/_tree.pyx":145
+/* "sklearn/tree/_tree.pyx":179
  *         self.reset()
  * 
  *     cdef void reset(self):             # <<<<<<<<<<<<<<
@@ -2110,22 +2185,93 @@ static void __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init(struct
  */
 
 static void __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_reset(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self) {
+  int __pyx_v_n_outputs;
+  int *__pyx_v_n_classes;
+  int __pyx_v_label_count_stride;
+  int *__pyx_v_label_count_init;
+  int *__pyx_v_label_count_left;
+  int *__pyx_v_label_count_right;
+  int __pyx_v_k;
   int __pyx_v_c;
   __Pyx_RefNannyDeclarations
   int __pyx_t_1;
+  int __pyx_t_2;
   __Pyx_RefNannySetupContext("reset", 0);
 
-  /* "sklearn/tree/_tree.pyx":148
+  /* "sklearn/tree/_tree.pyx":182
  *         """Reset label_counts by setting `label_count_left to zero
  *         and copying the init array into the right."""
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
+ */
+  __pyx_v_n_outputs = __pyx_v_self->n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":183
+ *         and copying the init array into the right."""
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes             # <<<<<<<<<<<<<<
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_init = self.label_count_init
+ */
+  __pyx_v_n_classes = __pyx_v_self->n_classes;
+
+  /* "sklearn/tree/_tree.pyx":184
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_init = self.label_count_init
+ *         cdef int* label_count_left = self.label_count_left
+ */
+  __pyx_v_label_count_stride = __pyx_v_self->label_count_stride;
+
+  /* "sklearn/tree/_tree.pyx":185
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_init = self.label_count_init             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right
+ */
+  __pyx_v_label_count_init = __pyx_v_self->label_count_init;
+
+  /* "sklearn/tree/_tree.pyx":186
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_init = self.label_count_init
+ *         cdef int* label_count_left = self.label_count_left             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_right = self.label_count_right
+ * 
+ */
+  __pyx_v_label_count_left = __pyx_v_self->label_count_left;
+
+  /* "sklearn/tree/_tree.pyx":187
+ *         cdef int* label_count_init = self.label_count_init
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int k = 0
+ */
+  __pyx_v_label_count_right = __pyx_v_self->label_count_right;
+
+  /* "sklearn/tree/_tree.pyx":189
+ *         cdef int* label_count_right = self.label_count_right
+ * 
+ *         cdef int k = 0             # <<<<<<<<<<<<<<
+ *         cdef int c = 0
+ *         self.n_left = 0
+ */
+  __pyx_v_k = 0;
+
+  /* "sklearn/tree/_tree.pyx":190
+ * 
+ *         cdef int k = 0
  *         cdef int c = 0             # <<<<<<<<<<<<<<
  *         self.n_left = 0
  *         self.n_right = self.n_samples
  */
   __pyx_v_c = 0;
 
-  /* "sklearn/tree/_tree.pyx":149
- *         and copying the init array into the right."""
+  /* "sklearn/tree/_tree.pyx":191
+ *         cdef int k = 0
  *         cdef int c = 0
  *         self.n_left = 0             # <<<<<<<<<<<<<<
  *         self.n_right = self.n_samples
@@ -2133,164 +2279,266 @@ static void __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_reset(struct
  */
   __pyx_v_self->n_left = 0;
 
-  /* "sklearn/tree/_tree.pyx":150
+  /* "sklearn/tree/_tree.pyx":192
  *         cdef int c = 0
  *         self.n_left = 0
  *         self.n_right = self.n_samples             # <<<<<<<<<<<<<<
  * 
- *         for c from 0 <= c < self.n_classes:
+ *         for k from 0 <= k < n_outputs:
  */
   __pyx_v_self->n_right = __pyx_v_self->n_samples;
 
-  /* "sklearn/tree/_tree.pyx":152
+  /* "sklearn/tree/_tree.pyx":194
  *         self.n_right = self.n_samples
  * 
- *         for c from 0 <= c < self.n_classes:             # <<<<<<<<<<<<<<
- *             self.label_count_left[c] = 0
- *             self.label_count_right[c] = self.label_count_init[c]
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             for c from 0 <= c < n_classes[k]:
+ *                 label_count_left[k * label_count_stride + c] = 0
  */
-  __pyx_t_1 = __pyx_v_self->n_classes;
-  for (__pyx_v_c = 0; __pyx_v_c < __pyx_t_1; __pyx_v_c++) {
+  __pyx_t_1 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
 
-    /* "sklearn/tree/_tree.pyx":153
+    /* "sklearn/tree/_tree.pyx":195
  * 
- *         for c from 0 <= c < self.n_classes:
- *             self.label_count_left[c] = 0             # <<<<<<<<<<<<<<
- *             self.label_count_right[c] = self.label_count_init[c]
+ *         for k from 0 <= k < n_outputs:
+ *             for c from 0 <= c < n_classes[k]:             # <<<<<<<<<<<<<<
+ *                 label_count_left[k * label_count_stride + c] = 0
+ *                 label_count_right[k * label_count_stride + c] = label_count_init[k * label_count_stride + c]
+ */
+    __pyx_t_2 = (__pyx_v_n_classes[__pyx_v_k]);
+    for (__pyx_v_c = 0; __pyx_v_c < __pyx_t_2; __pyx_v_c++) {
+
+      /* "sklearn/tree/_tree.pyx":196
+ *         for k from 0 <= k < n_outputs:
+ *             for c from 0 <= c < n_classes[k]:
+ *                 label_count_left[k * label_count_stride + c] = 0             # <<<<<<<<<<<<<<
+ *                 label_count_right[k * label_count_stride + c] = label_count_init[k * label_count_stride + c]
  * 
  */
-    (__pyx_v_self->label_count_left[__pyx_v_c]) = 0;
+      (__pyx_v_label_count_left[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]) = 0;
 
-    /* "sklearn/tree/_tree.pyx":154
- *         for c from 0 <= c < self.n_classes:
- *             self.label_count_left[c] = 0
- *             self.label_count_right[c] = self.label_count_init[c]             # <<<<<<<<<<<<<<
+      /* "sklearn/tree/_tree.pyx":197
+ *             for c from 0 <= c < n_classes[k]:
+ *                 label_count_left[k * label_count_stride + c] = 0
+ *                 label_count_right[k * label_count_stride + c] = label_count_init[k * label_count_stride + c]             # <<<<<<<<<<<<<<
  * 
- *     cdef int update(self, int a, int b, DTYPE_t *y, int *X_argsorted_i,
+ *     cdef int update(self, int a,
  */
-    (__pyx_v_self->label_count_right[__pyx_v_c]) = (__pyx_v_self->label_count_init[__pyx_v_c]);
+      (__pyx_v_label_count_right[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]) = (__pyx_v_label_count_init[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]);
+    }
   }
 
   __Pyx_RefNannyFinishContext();
 }
 
-/* "sklearn/tree/_tree.pyx":156
- *             self.label_count_right[c] = self.label_count_init[c]
+/* "sklearn/tree/_tree.pyx":199
+ *                 label_count_right[k * label_count_stride + c] = label_count_init[k * label_count_stride + c]
  * 
- *     cdef int update(self, int a, int b, DTYPE_t *y, int *X_argsorted_i,             # <<<<<<<<<<<<<<
- *                     BOOL_t *sample_mask):
- *         """Update the criteria for each value in interval [a,b) (where a and b
+ *     cdef int update(self, int a,             # <<<<<<<<<<<<<<
+ *                           int b,
+ *                           DTYPE_t* y,
  */
 
-static int __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_update(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self, int __pyx_v_a, int __pyx_v_b, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, int *__pyx_v_X_argsorted_i, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask) {
-  int __pyx_v_c;
+static int __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_update(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self, int __pyx_v_a, int __pyx_v_b, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, int __pyx_v_y_stride, int *__pyx_v_X_argsorted_i, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask) {
+  int __pyx_v_n_outputs;
+  int __pyx_v_label_count_stride;
+  int *__pyx_v_label_count_left;
+  int *__pyx_v_label_count_right;
+  int __pyx_v_n_left;
+  int __pyx_v_n_right;
   int __pyx_v_idx;
+  int __pyx_v_k;
+  int __pyx_v_c;
   int __pyx_v_s;
   int __pyx_r;
   __Pyx_RefNannyDeclarations
   int __pyx_t_1;
   int __pyx_t_2;
   int __pyx_t_3;
+  int __pyx_t_4;
   __Pyx_RefNannySetupContext("update", 0);
 
-  /* "sklearn/tree/_tree.pyx":162
- *         cdef int c
+  /* "sklearn/tree/_tree.pyx":207
+ *         """Update the criteria for each value in interval [a,b) (where a and b
+ *            are indices in `X_argsorted_i`)."""
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_left = self.label_count_left
+ */
+  __pyx_v_n_outputs = __pyx_v_self->n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":208
+ *            are indices in `X_argsorted_i`)."""
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int label_count_stride = self.label_count_stride             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right
+ */
+  __pyx_v_label_count_stride = __pyx_v_self->label_count_stride;
+
+  /* "sklearn/tree/_tree.pyx":209
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_left = self.label_count_left             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_right = self.label_count_right
+ *         cdef int n_left = self.n_left
+ */
+  __pyx_v_label_count_left = __pyx_v_self->label_count_left;
+
+  /* "sklearn/tree/_tree.pyx":210
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right             # <<<<<<<<<<<<<<
+ *         cdef int n_left = self.n_left
+ *         cdef int n_right = self.n_right
+ */
+  __pyx_v_label_count_right = __pyx_v_self->label_count_right;
+
+  /* "sklearn/tree/_tree.pyx":211
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right
+ *         cdef int n_left = self.n_left             # <<<<<<<<<<<<<<
+ *         cdef int n_right = self.n_right
+ * 
+ */
+  __pyx_v_n_left = __pyx_v_self->n_left;
+
+  /* "sklearn/tree/_tree.pyx":212
+ *         cdef int* label_count_right = self.label_count_right
+ *         cdef int n_left = self.n_left
+ *         cdef int n_right = self.n_right             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int idx, k, c, s
+ */
+  __pyx_v_n_right = __pyx_v_self->n_right;
+
+  /* "sklearn/tree/_tree.pyx":217
+ * 
  *         # post condition: all samples from [0:b) are on the left side
  *         for idx from a <= idx < b:             # <<<<<<<<<<<<<<
  *             s = X_argsorted_i[idx]
- *             if sample_mask[s] == 0:
+ * 
  */
   __pyx_t_1 = __pyx_v_b;
   for (__pyx_v_idx = __pyx_v_a; __pyx_v_idx < __pyx_t_1; __pyx_v_idx++) {
 
-    /* "sklearn/tree/_tree.pyx":163
+    /* "sklearn/tree/_tree.pyx":218
  *         # post condition: all samples from [0:b) are on the left side
  *         for idx from a <= idx < b:
  *             s = X_argsorted_i[idx]             # <<<<<<<<<<<<<<
+ * 
  *             if sample_mask[s] == 0:
- *                 continue
  */
     __pyx_v_s = (__pyx_v_X_argsorted_i[__pyx_v_idx]);
 
-    /* "sklearn/tree/_tree.pyx":164
- *         for idx from a <= idx < b:
+    /* "sklearn/tree/_tree.pyx":220
  *             s = X_argsorted_i[idx]
+ * 
  *             if sample_mask[s] == 0:             # <<<<<<<<<<<<<<
  *                 continue
- *             c = <int>(y[s])
+ * 
  */
     __pyx_t_2 = ((__pyx_v_sample_mask[__pyx_v_s]) == 0);
     if (__pyx_t_2) {
 
-      /* "sklearn/tree/_tree.pyx":165
- *             s = X_argsorted_i[idx]
+      /* "sklearn/tree/_tree.pyx":221
+ * 
  *             if sample_mask[s] == 0:
  *                 continue             # <<<<<<<<<<<<<<
- *             c = <int>(y[s])
- *             self.label_count_right[c] -= 1
+ * 
+ *             for k from 0 <= k < n_outputs:
  */
       goto __pyx_L3_continue;
       goto __pyx_L5;
     }
     __pyx_L5:;
 
-    /* "sklearn/tree/_tree.pyx":166
- *             if sample_mask[s] == 0:
+    /* "sklearn/tree/_tree.pyx":223
  *                 continue
- *             c = <int>(y[s])             # <<<<<<<<<<<<<<
- *             self.label_count_right[c] -= 1
- *             self.label_count_left[c] += 1
+ * 
+ *             for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *                 c = <int>y[s * y_stride + k]
+ *                 label_count_right[k * label_count_stride + c] -= 1
  */
-    __pyx_v_c = ((int)(__pyx_v_y[__pyx_v_s]));
+    __pyx_t_3 = __pyx_v_n_outputs;
+    for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_3; __pyx_v_k++) {
 
-    /* "sklearn/tree/_tree.pyx":167
- *                 continue
- *             c = <int>(y[s])
- *             self.label_count_right[c] -= 1             # <<<<<<<<<<<<<<
- *             self.label_count_left[c] += 1
- *             self.n_right -= 1
+      /* "sklearn/tree/_tree.pyx":224
+ * 
+ *             for k from 0 <= k < n_outputs:
+ *                 c = <int>y[s * y_stride + k]             # <<<<<<<<<<<<<<
+ *                 label_count_right[k * label_count_stride + c] -= 1
+ *                 label_count_left[k * label_count_stride + c] += 1
  */
-    __pyx_t_3 = __pyx_v_c;
-    (__pyx_v_self->label_count_right[__pyx_t_3]) = ((__pyx_v_self->label_count_right[__pyx_t_3]) - 1);
+      __pyx_v_c = ((int)(__pyx_v_y[((__pyx_v_s * __pyx_v_y_stride) + __pyx_v_k)]));
 
-    /* "sklearn/tree/_tree.pyx":168
- *             c = <int>(y[s])
- *             self.label_count_right[c] -= 1
- *             self.label_count_left[c] += 1             # <<<<<<<<<<<<<<
- *             self.n_right -= 1
- *             self.n_left += 1
+      /* "sklearn/tree/_tree.pyx":225
+ *             for k from 0 <= k < n_outputs:
+ *                 c = <int>y[s * y_stride + k]
+ *                 label_count_right[k * label_count_stride + c] -= 1             # <<<<<<<<<<<<<<
+ *                 label_count_left[k * label_count_stride + c] += 1
+ * 
  */
-    __pyx_t_3 = __pyx_v_c;
-    (__pyx_v_self->label_count_left[__pyx_t_3]) = ((__pyx_v_self->label_count_left[__pyx_t_3]) + 1);
+      __pyx_t_4 = ((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c);
+      (__pyx_v_label_count_right[__pyx_t_4]) = ((__pyx_v_label_count_right[__pyx_t_4]) - 1);
 
-    /* "sklearn/tree/_tree.pyx":169
- *             self.label_count_right[c] -= 1
- *             self.label_count_left[c] += 1
- *             self.n_right -= 1             # <<<<<<<<<<<<<<
- *             self.n_left += 1
+      /* "sklearn/tree/_tree.pyx":226
+ *                 c = <int>y[s * y_stride + k]
+ *                 label_count_right[k * label_count_stride + c] -= 1
+ *                 label_count_left[k * label_count_stride + c] += 1             # <<<<<<<<<<<<<<
  * 
+ *             n_left += 1
  */
-    __pyx_v_self->n_right = (__pyx_v_self->n_right - 1);
+      __pyx_t_4 = ((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c);
+      (__pyx_v_label_count_left[__pyx_t_4]) = ((__pyx_v_label_count_left[__pyx_t_4]) + 1);
+    }
 
-    /* "sklearn/tree/_tree.pyx":170
- *             self.label_count_left[c] += 1
- *             self.n_right -= 1
- *             self.n_left += 1             # <<<<<<<<<<<<<<
+    /* "sklearn/tree/_tree.pyx":228
+ *                 label_count_left[k * label_count_stride + c] += 1
+ * 
+ *             n_left += 1             # <<<<<<<<<<<<<<
+ *             n_right -=1
  * 
- *         return self.n_left
  */
-    __pyx_v_self->n_left = (__pyx_v_self->n_left + 1);
+    __pyx_v_n_left = (__pyx_v_n_left + 1);
+
+    /* "sklearn/tree/_tree.pyx":229
+ * 
+ *             n_left += 1
+ *             n_right -=1             # <<<<<<<<<<<<<<
+ * 
+ *         self.n_left = n_left
+ */
+    __pyx_v_n_right = (__pyx_v_n_right - 1);
     __pyx_L3_continue:;
   }
 
-  /* "sklearn/tree/_tree.pyx":172
- *             self.n_left += 1
+  /* "sklearn/tree/_tree.pyx":231
+ *             n_right -=1
+ * 
+ *         self.n_left = n_left             # <<<<<<<<<<<<<<
+ *         self.n_right = n_right
+ * 
+ */
+  __pyx_v_self->n_left = __pyx_v_n_left;
+
+  /* "sklearn/tree/_tree.pyx":232
+ * 
+ *         self.n_left = n_left
+ *         self.n_right = n_right             # <<<<<<<<<<<<<<
+ * 
+ *         return n_left
+ */
+  __pyx_v_self->n_right = __pyx_v_n_right;
+
+  /* "sklearn/tree/_tree.pyx":234
+ *         self.n_right = n_right
  * 
- *         return self.n_left             # <<<<<<<<<<<<<<
+ *         return n_left             # <<<<<<<<<<<<<<
  * 
  *     cdef double eval(self):
  */
-  __pyx_r = __pyx_v_self->n_left;
+  __pyx_r = __pyx_v_n_left;
   goto __pyx_L0;
 
   __pyx_r = 0;
@@ -2299,8 +2547,8 @@ static int __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_update(struct
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":174
- *         return self.n_left
+/* "sklearn/tree/_tree.pyx":236
+ *         return n_left
  * 
  *     cdef double eval(self):             # <<<<<<<<<<<<<<
  *         pass
@@ -2317,35 +2565,55 @@ static double __pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_eval(CYTHO
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":177
+/* "sklearn/tree/_tree.pyx":239
  *         pass
  * 
  *     cpdef np.ndarray init_value(self):             # <<<<<<<<<<<<<<
- *         return self.ndarray_label_count_init
- * 
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes
  */
 
-static PyObject *__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_3init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_5init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
 static PyArrayObject *__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init_value(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self, int __pyx_skip_dispatch) {
+  int __pyx_v_n_outputs;
+  int *__pyx_v_n_classes;
+  int __pyx_v_label_count_stride;
+  int *__pyx_v_label_count_init;
+  PyArrayObject *__pyx_v_value = 0;
+  int __pyx_v_k;
+  int __pyx_v_c;
+  __Pyx_LocalBuf_ND __pyx_pybuffernd_value;
+  __Pyx_Buffer __pyx_pybuffer_value;
   PyArrayObject *__pyx_r = NULL;
   __Pyx_RefNannyDeclarations
   PyObject *__pyx_t_1 = NULL;
   PyObject *__pyx_t_2 = NULL;
+  PyObject *__pyx_t_3 = NULL;
+  PyObject *__pyx_t_4 = NULL;
+  PyArrayObject *__pyx_t_5 = NULL;
+  int __pyx_t_6;
+  int __pyx_t_7;
+  int __pyx_t_8;
+  int __pyx_t_9;
   int __pyx_lineno = 0;
   const char *__pyx_filename = NULL;
   int __pyx_clineno = 0;
   __Pyx_RefNannySetupContext("init_value", 0);
+  __pyx_pybuffer_value.pybuffer.buf = NULL;
+  __pyx_pybuffer_value.refcount = 0;
+  __pyx_pybuffernd_value.data = NULL;
+  __pyx_pybuffernd_value.rcbuffer = &__pyx_pybuffer_value;
   /* Check if called by wrapper */
   if (unlikely(__pyx_skip_dispatch)) ;
   /* Check if overriden in Python */
   else if (unlikely(Py_TYPE(((PyObject *)__pyx_v_self))->tp_dictoffset != 0)) {
-    __pyx_t_1 = PyObject_GetAttr(((PyObject *)__pyx_v_self), __pyx_n_s__init_value); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 177; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_1 = PyObject_GetAttr(((PyObject *)__pyx_v_self), __pyx_n_s__init_value); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 239; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     __Pyx_GOTREF(__pyx_t_1);
-    if (!PyCFunction_Check(__pyx_t_1) || (PyCFunction_GET_FUNCTION(__pyx_t_1) != (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_3init_value)) {
+    if (!PyCFunction_Check(__pyx_t_1) || (PyCFunction_GET_FUNCTION(__pyx_t_1) != (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_5init_value)) {
       __Pyx_XDECREF(((PyObject *)__pyx_r));
-      __pyx_t_2 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_empty_tuple), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 177; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      __pyx_t_2 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_empty_tuple), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 239; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
       __Pyx_GOTREF(__pyx_t_2);
-      if (!(likely(((__pyx_t_2) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_2, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 177; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      if (!(likely(((__pyx_t_2) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_2, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 239; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
       __pyx_r = ((PyArrayObject *)__pyx_t_2);
       __pyx_t_2 = 0;
       __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
@@ -2354,17 +2622,139 @@ static PyArrayObject *__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_in
     __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
   }
 
-  /* "sklearn/tree/_tree.pyx":178
+  /* "sklearn/tree/_tree.pyx":240
  * 
  *     cpdef np.ndarray init_value(self):
- *         return self.ndarray_label_count_init             # <<<<<<<<<<<<<<
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
+ */
+  __pyx_v_n_outputs = __pyx_v_self->n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":241
+ *     cpdef np.ndarray init_value(self):
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes             # <<<<<<<<<<<<<<
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_init = self.label_count_init
+ */
+  __pyx_v_n_classes = __pyx_v_self->n_classes;
+
+  /* "sklearn/tree/_tree.pyx":242
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_init = self.label_count_init
+ * 
+ */
+  __pyx_v_label_count_stride = __pyx_v_self->label_count_stride;
+
+  /* "sklearn/tree/_tree.pyx":243
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_init = self.label_count_init             # <<<<<<<<<<<<<<
+ * 
+ *         cdef np.ndarray[DTYPE_t, ndim=2] value = np.zeros((n_outputs, label_count_stride), dtype=DTYPE)
+ */
+  __pyx_v_label_count_init = __pyx_v_self->label_count_init;
+
+  /* "sklearn/tree/_tree.pyx":245
+ *         cdef int* label_count_init = self.label_count_init
+ * 
+ *         cdef np.ndarray[DTYPE_t, ndim=2] value = np.zeros((n_outputs, label_count_stride), dtype=DTYPE)             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int k, c
+ */
+  __pyx_t_1 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  __pyx_t_2 = PyObject_GetAttr(__pyx_t_1, __pyx_n_s__zeros); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_2);
+  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
+  __pyx_t_1 = PyInt_FromLong(__pyx_v_n_outputs); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  __pyx_t_3 = PyInt_FromLong(__pyx_v_label_count_stride); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_3);
+  __pyx_t_4 = PyTuple_New(2); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_4);
+  PyTuple_SET_ITEM(__pyx_t_4, 0, __pyx_t_1);
+  __Pyx_GIVEREF(__pyx_t_1);
+  PyTuple_SET_ITEM(__pyx_t_4, 1, __pyx_t_3);
+  __Pyx_GIVEREF(__pyx_t_3);
+  __pyx_t_1 = 0;
+  __pyx_t_3 = 0;
+  __pyx_t_3 = PyTuple_New(1); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_3);
+  PyTuple_SET_ITEM(__pyx_t_3, 0, ((PyObject *)__pyx_t_4));
+  __Pyx_GIVEREF(((PyObject *)__pyx_t_4));
+  __pyx_t_4 = 0;
+  __pyx_t_4 = PyDict_New(); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(((PyObject *)__pyx_t_4));
+  __pyx_t_1 = __Pyx_GetName(__pyx_m, __pyx_n_s__DTYPE); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  if (PyDict_SetItem(__pyx_t_4, ((PyObject *)__pyx_n_s__dtype), __pyx_t_1) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
+  __pyx_t_1 = PyObject_Call(__pyx_t_2, ((PyObject *)__pyx_t_3), ((PyObject *)__pyx_t_4)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+  __Pyx_DECREF(((PyObject *)__pyx_t_3)); __pyx_t_3 = 0;
+  __Pyx_DECREF(((PyObject *)__pyx_t_4)); __pyx_t_4 = 0;
+  if (!(likely(((__pyx_t_1) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_1, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_5 = ((PyArrayObject *)__pyx_t_1);
+  {
+    __Pyx_BufFmt_StackElem __pyx_stack[1];
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_value.rcbuffer->pybuffer, (PyObject*)__pyx_t_5, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_STRIDES| PyBUF_WRITABLE, 2, 0, __pyx_stack) == -1)) {
+      __pyx_v_value = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None); __pyx_pybuffernd_value.rcbuffer->pybuffer.buf = NULL;
+      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    } else {__pyx_pybuffernd_value.diminfo[0].strides = __pyx_pybuffernd_value.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_value.diminfo[0].shape = __pyx_pybuffernd_value.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_value.diminfo[1].strides = __pyx_pybuffernd_value.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_value.diminfo[1].shape = __pyx_pybuffernd_value.rcbuffer->pybuffer.shape[1];
+    }
+  }
+  __pyx_t_5 = 0;
+  __pyx_v_value = ((PyArrayObject *)__pyx_t_1);
+  __pyx_t_1 = 0;
+
+  /* "sklearn/tree/_tree.pyx":249
+ *         cdef int k, c
+ * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             for c from 0 <= c < n_classes[k]:
+ *                 value[k, c] = <DTYPE_t> (label_count_init[k * label_count_stride + c])
+ */
+  __pyx_t_6 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_6; __pyx_v_k++) {
+
+    /* "sklearn/tree/_tree.pyx":250
+ * 
+ *         for k from 0 <= k < n_outputs:
+ *             for c from 0 <= c < n_classes[k]:             # <<<<<<<<<<<<<<
+ *                 value[k, c] = <DTYPE_t> (label_count_init[k * label_count_stride + c])
+ * 
+ */
+    __pyx_t_7 = (__pyx_v_n_classes[__pyx_v_k]);
+    for (__pyx_v_c = 0; __pyx_v_c < __pyx_t_7; __pyx_v_c++) {
+
+      /* "sklearn/tree/_tree.pyx":251
+ *         for k from 0 <= k < n_outputs:
+ *             for c from 0 <= c < n_classes[k]:
+ *                 value[k, c] = <DTYPE_t> (label_count_init[k * label_count_stride + c])             # <<<<<<<<<<<<<<
+ * 
+ *         return value
+ */
+      __pyx_t_8 = __pyx_v_k;
+      __pyx_t_9 = __pyx_v_c;
+      *__Pyx_BufPtrStrided2d(__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_pybuffernd_value.rcbuffer->pybuffer.buf, __pyx_t_8, __pyx_pybuffernd_value.diminfo[0].strides, __pyx_t_9, __pyx_pybuffernd_value.diminfo[1].strides) = ((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t)(__pyx_v_label_count_init[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]));
+    }
+  }
+
+  /* "sklearn/tree/_tree.pyx":253
+ *                 value[k, c] = <DTYPE_t> (label_count_init[k * label_count_stride + c])
  * 
+ *         return value             # <<<<<<<<<<<<<<
  * 
+ * cdef class Gini(ClassificationCriterion):
  */
   __Pyx_XDECREF(((PyObject *)__pyx_r));
-  if (!(likely(((__pyx_v_self->ndarray_label_count_init) == Py_None) || likely(__Pyx_TypeTest(__pyx_v_self->ndarray_label_count_init, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 178; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_INCREF(__pyx_v_self->ndarray_label_count_init);
-  __pyx_r = ((PyArrayObject *)__pyx_v_self->ndarray_label_count_init);
+  __Pyx_INCREF(((PyObject *)__pyx_v_value));
+  __pyx_r = ((PyArrayObject *)__pyx_v_value);
   goto __pyx_L0;
 
   __pyx_r = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None);
@@ -2372,34 +2762,44 @@ static PyArrayObject *__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_in
   __pyx_L1_error:;
   __Pyx_XDECREF(__pyx_t_1);
   __Pyx_XDECREF(__pyx_t_2);
+  __Pyx_XDECREF(__pyx_t_3);
+  __Pyx_XDECREF(__pyx_t_4);
+  { PyObject *__pyx_type, *__pyx_value, *__pyx_tb;
+    __Pyx_ErrFetch(&__pyx_type, &__pyx_value, &__pyx_tb);
+    __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_value.rcbuffer->pybuffer);
+  __Pyx_ErrRestore(__pyx_type, __pyx_value, __pyx_tb);}
   __Pyx_AddTraceback("sklearn.tree._tree.ClassificationCriterion.init_value", __pyx_clineno, __pyx_lineno, __pyx_filename);
   __pyx_r = 0;
+  goto __pyx_L2;
   __pyx_L0:;
+  __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_value.rcbuffer->pybuffer);
+  __pyx_L2:;
+  __Pyx_XDECREF((PyObject *)__pyx_v_value);
   __Pyx_XGIVEREF((PyObject *)__pyx_r);
   __Pyx_RefNannyFinishContext();
   return __pyx_r;
 }
 
 /* Python wrapper */
-static PyObject *__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_3init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
-static PyObject *__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_3init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused) {
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_5init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_5init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused) {
   PyObject *__pyx_r = 0;
   __Pyx_RefNannyDeclarations
   __Pyx_RefNannySetupContext("init_value (wrapper)", 0);
-  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_2init_value(((struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *)__pyx_v_self));
+  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_4init_value(((struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *)__pyx_v_self));
   __Pyx_RefNannyFinishContext();
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":177
+/* "sklearn/tree/_tree.pyx":239
  *         pass
  * 
  *     cpdef np.ndarray init_value(self):             # <<<<<<<<<<<<<<
- *         return self.ndarray_label_count_init
- * 
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes
  */
 
-static PyObject *__pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_2init_value(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self) {
+static PyObject *__pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_4init_value(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *__pyx_v_self) {
   PyObject *__pyx_r = NULL;
   __Pyx_RefNannyDeclarations
   PyObject *__pyx_t_1 = NULL;
@@ -2408,7 +2808,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_2init_
   int __pyx_clineno = 0;
   __Pyx_RefNannySetupContext("init_value", 0);
   __Pyx_XDECREF(__pyx_r);
-  __pyx_t_1 = ((PyObject *)((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_ClassificationCriterion *)__pyx_v_self->__pyx_base.__pyx_vtab)->__pyx_base.init_value(((struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *)__pyx_v_self), 1)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 177; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = ((PyObject *)((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_ClassificationCriterion *)__pyx_v_self->__pyx_base.__pyx_vtab)->__pyx_base.init_value(((struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *)__pyx_v_self), 1)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 239; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
   __pyx_r = __pyx_t_1;
   __pyx_t_1 = 0;
@@ -2426,211 +2826,303 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_23ClassificationCriterion_2init_
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":188
+/* "sklearn/tree/_tree.pyx":262
  *     """
  * 
  *     cdef double eval(self):             # <<<<<<<<<<<<<<
  *         """Returns Gini index of left branch + Gini index of right branch. """
- *         cdef double n_left = <double> self.n_left
+ *         cdef int n_samples = self.n_samples
  */
 
 static double __pyx_f_7sklearn_4tree_5_tree_4Gini_eval(struct __pyx_obj_7sklearn_4tree_5_tree_Gini *__pyx_v_self) {
+  int __pyx_v_n_samples;
+  int __pyx_v_n_outputs;
+  int *__pyx_v_n_classes;
+  int __pyx_v_label_count_stride;
+  int *__pyx_v_label_count_left;
+  int *__pyx_v_label_count_right;
   double __pyx_v_n_left;
   double __pyx_v_n_right;
+  double __pyx_v_total;
   double __pyx_v_H_left;
   double __pyx_v_H_right;
   int __pyx_v_k;
+  int __pyx_v_c;
   int __pyx_v_count_left;
   int __pyx_v_count_right;
   double __pyx_r;
   __Pyx_RefNannyDeclarations
   int __pyx_t_1;
   int __pyx_t_2;
+  int __pyx_t_3;
   __Pyx_RefNannySetupContext("eval", 0);
 
-  /* "sklearn/tree/_tree.pyx":190
+  /* "sklearn/tree/_tree.pyx":264
  *     cdef double eval(self):
  *         """Returns Gini index of left branch + Gini index of right branch. """
+ *         cdef int n_samples = self.n_samples             # <<<<<<<<<<<<<<
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes
+ */
+  __pyx_v_n_samples = __pyx_v_self->__pyx_base.n_samples;
+
+  /* "sklearn/tree/_tree.pyx":265
+ *         """Returns Gini index of left branch + Gini index of right branch. """
+ *         cdef int n_samples = self.n_samples
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
+ */
+  __pyx_v_n_outputs = __pyx_v_self->__pyx_base.n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":266
+ *         cdef int n_samples = self.n_samples
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes             # <<<<<<<<<<<<<<
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_left = self.label_count_left
+ */
+  __pyx_v_n_classes = __pyx_v_self->__pyx_base.n_classes;
+
+  /* "sklearn/tree/_tree.pyx":267
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right
+ */
+  __pyx_v_label_count_stride = __pyx_v_self->__pyx_base.label_count_stride;
+
+  /* "sklearn/tree/_tree.pyx":268
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_left = self.label_count_left             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_right = self.label_count_right
+ *         cdef double n_left = <double> self.n_left
+ */
+  __pyx_v_label_count_left = __pyx_v_self->__pyx_base.label_count_left;
+
+  /* "sklearn/tree/_tree.pyx":269
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right             # <<<<<<<<<<<<<<
+ *         cdef double n_left = <double> self.n_left
+ *         cdef double n_right = <double> self.n_right
+ */
+  __pyx_v_label_count_right = __pyx_v_self->__pyx_base.label_count_right;
+
+  /* "sklearn/tree/_tree.pyx":270
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right
  *         cdef double n_left = <double> self.n_left             # <<<<<<<<<<<<<<
  *         cdef double n_right = <double> self.n_right
- *         cdef double H_left = n_left * n_left
+ * 
  */
   __pyx_v_n_left = ((double)__pyx_v_self->__pyx_base.n_left);
 
-  /* "sklearn/tree/_tree.pyx":191
- *         """Returns Gini index of left branch + Gini index of right branch. """
+  /* "sklearn/tree/_tree.pyx":271
+ *         cdef int* label_count_right = self.label_count_right
  *         cdef double n_left = <double> self.n_left
  *         cdef double n_right = <double> self.n_right             # <<<<<<<<<<<<<<
- *         cdef double H_left = n_left * n_left
- *         cdef double H_right = n_right * n_right
+ * 
+ *         cdef double total = 0.0
  */
   __pyx_v_n_right = ((double)__pyx_v_self->__pyx_base.n_right);
 
-  /* "sklearn/tree/_tree.pyx":192
- *         cdef double n_left = <double> self.n_left
+  /* "sklearn/tree/_tree.pyx":273
  *         cdef double n_right = <double> self.n_right
- *         cdef double H_left = n_left * n_left             # <<<<<<<<<<<<<<
- *         cdef double H_right = n_right * n_right
- *         cdef int k, count_left, count_right
+ * 
+ *         cdef double total = 0.0             # <<<<<<<<<<<<<<
+ *         cdef double H_left
+ *         cdef double H_right
  */
-  __pyx_v_H_left = (__pyx_v_n_left * __pyx_v_n_left);
+  __pyx_v_total = 0.0;
 
-  /* "sklearn/tree/_tree.pyx":193
- *         cdef double n_right = <double> self.n_right
- *         cdef double H_left = n_left * n_left
- *         cdef double H_right = n_right * n_right             # <<<<<<<<<<<<<<
- *         cdef int k, count_left, count_right
+  /* "sklearn/tree/_tree.pyx":278
+ *         cdef int k, c, count_left, count_right
  * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             H_left = n_left * n_left
+ *             H_right = n_right * n_right
  */
-  __pyx_v_H_right = (__pyx_v_n_right * __pyx_v_n_right);
+  __pyx_t_1 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
 
-  /* "sklearn/tree/_tree.pyx":196
- *         cdef int k, count_left, count_right
+    /* "sklearn/tree/_tree.pyx":279
+ * 
+ *         for k from 0 <= k < n_outputs:
+ *             H_left = n_left * n_left             # <<<<<<<<<<<<<<
+ *             H_right = n_right * n_right
  * 
- *         for k from 0 <= k < self.n_classes:             # <<<<<<<<<<<<<<
- *             count_left = self.label_count_left[k]
- *             if count_left > 0:
  */
-  __pyx_t_1 = __pyx_v_self->__pyx_base.n_classes;
-  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
+    __pyx_v_H_left = (__pyx_v_n_left * __pyx_v_n_left);
 
-    /* "sklearn/tree/_tree.pyx":197
+    /* "sklearn/tree/_tree.pyx":280
+ *         for k from 0 <= k < n_outputs:
+ *             H_left = n_left * n_left
+ *             H_right = n_right * n_right             # <<<<<<<<<<<<<<
  * 
- *         for k from 0 <= k < self.n_classes:
- *             count_left = self.label_count_left[k]             # <<<<<<<<<<<<<<
- *             if count_left > 0:
- *                 H_left -= (count_left * count_left)
+ *             for c from 0 <= c < n_classes[k]:
  */
-    __pyx_v_count_left = (__pyx_v_self->__pyx_base.label_count_left[__pyx_v_k]);
+    __pyx_v_H_right = (__pyx_v_n_right * __pyx_v_n_right);
 
-    /* "sklearn/tree/_tree.pyx":198
- *         for k from 0 <= k < self.n_classes:
- *             count_left = self.label_count_left[k]
- *             if count_left > 0:             # <<<<<<<<<<<<<<
- *                 H_left -= (count_left * count_left)
- *             count_right = self.label_count_right[k]
+    /* "sklearn/tree/_tree.pyx":282
+ *             H_right = n_right * n_right
+ * 
+ *             for c from 0 <= c < n_classes[k]:             # <<<<<<<<<<<<<<
+ *                 count_left = label_count_left[k * label_count_stride + c]
+ *                 if count_left > 0:
  */
-    __pyx_t_2 = (__pyx_v_count_left > 0);
-    if (__pyx_t_2) {
+    __pyx_t_2 = (__pyx_v_n_classes[__pyx_v_k]);
+    for (__pyx_v_c = 0; __pyx_v_c < __pyx_t_2; __pyx_v_c++) {
 
-      /* "sklearn/tree/_tree.pyx":199
- *             count_left = self.label_count_left[k]
- *             if count_left > 0:
- *                 H_left -= (count_left * count_left)             # <<<<<<<<<<<<<<
- *             count_right = self.label_count_right[k]
- *             if count_right > 0:
+      /* "sklearn/tree/_tree.pyx":283
+ * 
+ *             for c from 0 <= c < n_classes[k]:
+ *                 count_left = label_count_left[k * label_count_stride + c]             # <<<<<<<<<<<<<<
+ *                 if count_left > 0:
+ *                     H_left -= (count_left * count_left)
  */
-      __pyx_v_H_left = (__pyx_v_H_left - (__pyx_v_count_left * __pyx_v_count_left));
-      goto __pyx_L5;
-    }
-    __pyx_L5:;
+      __pyx_v_count_left = (__pyx_v_label_count_left[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]);
 
-    /* "sklearn/tree/_tree.pyx":200
- *             if count_left > 0:
- *                 H_left -= (count_left * count_left)
- *             count_right = self.label_count_right[k]             # <<<<<<<<<<<<<<
- *             if count_right > 0:
- *                 H_right -= (count_right * count_right)
+      /* "sklearn/tree/_tree.pyx":284
+ *             for c from 0 <= c < n_classes[k]:
+ *                 count_left = label_count_left[k * label_count_stride + c]
+ *                 if count_left > 0:             # <<<<<<<<<<<<<<
+ *                     H_left -= (count_left * count_left)
+ * 
  */
-    __pyx_v_count_right = (__pyx_v_self->__pyx_base.label_count_right[__pyx_v_k]);
+      __pyx_t_3 = (__pyx_v_count_left > 0);
+      if (__pyx_t_3) {
 
-    /* "sklearn/tree/_tree.pyx":201
- *                 H_left -= (count_left * count_left)
- *             count_right = self.label_count_right[k]
- *             if count_right > 0:             # <<<<<<<<<<<<<<
- *                 H_right -= (count_right * count_right)
+        /* "sklearn/tree/_tree.pyx":285
+ *                 count_left = label_count_left[k * label_count_stride + c]
+ *                 if count_left > 0:
+ *                     H_left -= (count_left * count_left)             # <<<<<<<<<<<<<<
  * 
+ *                 count_right = label_count_right[k * label_count_stride + c]
  */
-    __pyx_t_2 = (__pyx_v_count_right > 0);
-    if (__pyx_t_2) {
+        __pyx_v_H_left = (__pyx_v_H_left - (__pyx_v_count_left * __pyx_v_count_left));
+        goto __pyx_L7;
+      }
+      __pyx_L7:;
 
-      /* "sklearn/tree/_tree.pyx":202
- *             count_right = self.label_count_right[k]
- *             if count_right > 0:
- *                 H_right -= (count_right * count_right)             # <<<<<<<<<<<<<<
+      /* "sklearn/tree/_tree.pyx":287
+ *                     H_left -= (count_left * count_left)
  * 
- *         if n_left == 0:
+ *                 count_right = label_count_right[k * label_count_stride + c]             # <<<<<<<<<<<<<<
+ *                 if count_right > 0:
+ *                     H_right -= (count_right * count_right)
  */
-      __pyx_v_H_right = (__pyx_v_H_right - (__pyx_v_count_right * __pyx_v_count_right));
-      goto __pyx_L6;
+      __pyx_v_count_right = (__pyx_v_label_count_right[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]);
+
+      /* "sklearn/tree/_tree.pyx":288
+ * 
+ *                 count_right = label_count_right[k * label_count_stride + c]
+ *                 if count_right > 0:             # <<<<<<<<<<<<<<
+ *                     H_right -= (count_right * count_right)
+ * 
+ */
+      __pyx_t_3 = (__pyx_v_count_right > 0);
+      if (__pyx_t_3) {
+
+        /* "sklearn/tree/_tree.pyx":289
+ *                 count_right = label_count_right[k * label_count_stride + c]
+ *                 if count_right > 0:
+ *                     H_right -= (count_right * count_right)             # <<<<<<<<<<<<<<
+ * 
+ *             if n_left == 0:
+ */
+        __pyx_v_H_right = (__pyx_v_H_right - (__pyx_v_count_right * __pyx_v_count_right));
+        goto __pyx_L8;
+      }
+      __pyx_L8:;
     }
-    __pyx_L6:;
-  }
 
-  /* "sklearn/tree/_tree.pyx":204
- *                 H_right -= (count_right * count_right)
+    /* "sklearn/tree/_tree.pyx":291
+ *                     H_right -= (count_right * count_right)
  * 
- *         if n_left == 0:             # <<<<<<<<<<<<<<
- *             H_left = 0
- *         else:
+ *             if n_left == 0:             # <<<<<<<<<<<<<<
+ *                 H_left = 0
+ *             else:
  */
-  __pyx_t_2 = (__pyx_v_n_left == 0.0);
-  if (__pyx_t_2) {
+    __pyx_t_3 = (__pyx_v_n_left == 0.0);
+    if (__pyx_t_3) {
 
-    /* "sklearn/tree/_tree.pyx":205
+      /* "sklearn/tree/_tree.pyx":292
  * 
- *         if n_left == 0:
- *             H_left = 0             # <<<<<<<<<<<<<<
- *         else:
- *             H_left /= n_left
+ *             if n_left == 0:
+ *                 H_left = 0             # <<<<<<<<<<<<<<
+ *             else:
+ *                 H_left /= n_left
  */
-    __pyx_v_H_left = 0.0;
-    goto __pyx_L7;
-  }
-  /*else*/ {
+      __pyx_v_H_left = 0.0;
+      goto __pyx_L9;
+    }
+    /*else*/ {
 
-    /* "sklearn/tree/_tree.pyx":207
- *             H_left = 0
- *         else:
- *             H_left /= n_left             # <<<<<<<<<<<<<<
+      /* "sklearn/tree/_tree.pyx":294
+ *                 H_left = 0
+ *             else:
+ *                 H_left /= n_left             # <<<<<<<<<<<<<<
  * 
- *         if n_right == 0:
+ *             if n_right == 0:
  */
-    __pyx_v_H_left = (__pyx_v_H_left / __pyx_v_n_left);
-  }
-  __pyx_L7:;
+      __pyx_v_H_left = (__pyx_v_H_left / __pyx_v_n_left);
+    }
+    __pyx_L9:;
 
-  /* "sklearn/tree/_tree.pyx":209
- *             H_left /= n_left
+    /* "sklearn/tree/_tree.pyx":296
+ *                 H_left /= n_left
  * 
- *         if n_right == 0:             # <<<<<<<<<<<<<<
- *             H_right = 0
- *         else:
+ *             if n_right == 0:             # <<<<<<<<<<<<<<
+ *                 H_right = 0
+ *             else:
  */
-  __pyx_t_2 = (__pyx_v_n_right == 0.0);
-  if (__pyx_t_2) {
+    __pyx_t_3 = (__pyx_v_n_right == 0.0);
+    if (__pyx_t_3) {
 
-    /* "sklearn/tree/_tree.pyx":210
+      /* "sklearn/tree/_tree.pyx":297
  * 
- *         if n_right == 0:
- *             H_right = 0             # <<<<<<<<<<<<<<
- *         else:
- *             H_right /= n_right
+ *             if n_right == 0:
+ *                 H_right = 0             # <<<<<<<<<<<<<<
+ *             else:
+ *                 H_right /= n_right
  */
-    __pyx_v_H_right = 0.0;
-    goto __pyx_L8;
-  }
-  /*else*/ {
+      __pyx_v_H_right = 0.0;
+      goto __pyx_L10;
+    }
+    /*else*/ {
 
-    /* "sklearn/tree/_tree.pyx":212
- *             H_right = 0
- *         else:
- *             H_right /= n_right             # <<<<<<<<<<<<<<
+      /* "sklearn/tree/_tree.pyx":299
+ *                 H_right = 0
+ *             else:
+ *                 H_right /= n_right             # <<<<<<<<<<<<<<
+ * 
+ *             total += (H_left + H_right)
+ */
+      __pyx_v_H_right = (__pyx_v_H_right / __pyx_v_n_right);
+    }
+    __pyx_L10:;
+
+    /* "sklearn/tree/_tree.pyx":301
+ *                 H_right /= n_right
+ * 
+ *             total += (H_left + H_right)             # <<<<<<<<<<<<<<
  * 
- *         return (H_left + H_right) / self.n_samples
+ *         return total / (n_samples * n_outputs)
  */
-    __pyx_v_H_right = (__pyx_v_H_right / __pyx_v_n_right);
+    __pyx_v_total = (__pyx_v_total + (__pyx_v_H_left + __pyx_v_H_right));
   }
-  __pyx_L8:;
 
-  /* "sklearn/tree/_tree.pyx":214
- *             H_right /= n_right
+  /* "sklearn/tree/_tree.pyx":303
+ *             total += (H_left + H_right)
  * 
- *         return (H_left + H_right) / self.n_samples             # <<<<<<<<<<<<<<
+ *         return total / (n_samples * n_outputs)             # <<<<<<<<<<<<<<
  * 
  * 
  */
-  __pyx_r = ((__pyx_v_H_left + __pyx_v_H_right) / __pyx_v_self->__pyx_base.n_samples);
+  __pyx_r = (__pyx_v_total / (__pyx_v_n_samples * __pyx_v_n_outputs));
   goto __pyx_L0;
 
   __pyx_r = 0;
@@ -2639,145 +3131,237 @@ static double __pyx_f_7sklearn_4tree_5_tree_4Gini_eval(struct __pyx_obj_7sklearn
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":223
+/* "sklearn/tree/_tree.pyx":312
  *     """
  * 
  *     cdef double eval(self):             # <<<<<<<<<<<<<<
  *         """Returns Entropy of left branch + Entropy index of right branch. """
- *         cdef double H_left = 0.0
+ *         cdef int n_samples = self.n_samples
  */
 
 static double __pyx_f_7sklearn_4tree_5_tree_7Entropy_eval(struct __pyx_obj_7sklearn_4tree_5_tree_Entropy *__pyx_v_self) {
+  int __pyx_v_n_samples;
+  int __pyx_v_n_outputs;
+  int *__pyx_v_n_classes;
+  int __pyx_v_label_count_stride;
+  int *__pyx_v_label_count_left;
+  int *__pyx_v_label_count_right;
+  double __pyx_v_n_left;
+  double __pyx_v_n_right;
+  double __pyx_v_total;
   double __pyx_v_H_left;
   double __pyx_v_H_right;
   int __pyx_v_k;
+  int __pyx_v_c;
   double __pyx_v_e1;
   double __pyx_v_e2;
-  double __pyx_v_n_left;
-  double __pyx_v_n_right;
   double __pyx_r;
   __Pyx_RefNannyDeclarations
   int __pyx_t_1;
   int __pyx_t_2;
+  int __pyx_t_3;
   __Pyx_RefNannySetupContext("eval", 0);
 
-  /* "sklearn/tree/_tree.pyx":225
+  /* "sklearn/tree/_tree.pyx":314
  *     cdef double eval(self):
  *         """Returns Entropy of left branch + Entropy index of right branch. """
- *         cdef double H_left = 0.0             # <<<<<<<<<<<<<<
- *         cdef double H_right = 0.0
- *         cdef int k
+ *         cdef int n_samples = self.n_samples             # <<<<<<<<<<<<<<
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes
  */
-  __pyx_v_H_left = 0.0;
+  __pyx_v_n_samples = __pyx_v_self->__pyx_base.n_samples;
 
-  /* "sklearn/tree/_tree.pyx":226
+  /* "sklearn/tree/_tree.pyx":315
  *         """Returns Entropy of left branch + Entropy index of right branch. """
- *         cdef double H_left = 0.0
- *         cdef double H_right = 0.0             # <<<<<<<<<<<<<<
- *         cdef int k
- *         cdef double e1, e2
+ *         cdef int n_samples = self.n_samples
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
  */
-  __pyx_v_H_right = 0.0;
+  __pyx_v_n_outputs = __pyx_v_self->__pyx_base.n_outputs;
 
-  /* "sklearn/tree/_tree.pyx":229
- *         cdef int k
- *         cdef double e1, e2
+  /* "sklearn/tree/_tree.pyx":316
+ *         cdef int n_samples = self.n_samples
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes             # <<<<<<<<<<<<<<
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_left = self.label_count_left
+ */
+  __pyx_v_n_classes = __pyx_v_self->__pyx_base.n_classes;
+
+  /* "sklearn/tree/_tree.pyx":317
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right
+ */
+  __pyx_v_label_count_stride = __pyx_v_self->__pyx_base.label_count_stride;
+
+  /* "sklearn/tree/_tree.pyx":318
+ *         cdef int* n_classes = self.n_classes
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_left = self.label_count_left             # <<<<<<<<<<<<<<
+ *         cdef int* label_count_right = self.label_count_right
+ *         cdef double n_left = <double> self.n_left
+ */
+  __pyx_v_label_count_left = __pyx_v_self->__pyx_base.label_count_left;
+
+  /* "sklearn/tree/_tree.pyx":319
+ *         cdef int label_count_stride = self.label_count_stride
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right             # <<<<<<<<<<<<<<
+ *         cdef double n_left = <double> self.n_left
+ *         cdef double n_right = <double> self.n_right
+ */
+  __pyx_v_label_count_right = __pyx_v_self->__pyx_base.label_count_right;
+
+  /* "sklearn/tree/_tree.pyx":320
+ *         cdef int* label_count_left = self.label_count_left
+ *         cdef int* label_count_right = self.label_count_right
  *         cdef double n_left = <double> self.n_left             # <<<<<<<<<<<<<<
  *         cdef double n_right = <double> self.n_right
  * 
  */
   __pyx_v_n_left = ((double)__pyx_v_self->__pyx_base.n_left);
 
-  /* "sklearn/tree/_tree.pyx":230
- *         cdef double e1, e2
+  /* "sklearn/tree/_tree.pyx":321
+ *         cdef int* label_count_right = self.label_count_right
  *         cdef double n_left = <double> self.n_left
  *         cdef double n_right = <double> self.n_right             # <<<<<<<<<<<<<<
  * 
- *         for k from 0 <= k < self.n_classes:
+ *         cdef double total = 0.0
  */
   __pyx_v_n_right = ((double)__pyx_v_self->__pyx_base.n_right);
 
-  /* "sklearn/tree/_tree.pyx":232
+  /* "sklearn/tree/_tree.pyx":323
  *         cdef double n_right = <double> self.n_right
  * 
- *         for k from 0 <= k < self.n_classes:             # <<<<<<<<<<<<<<
- *             if self.label_count_left[k] > 0:
- *                 H_left -= ((self.label_count_left[k] / n_left)
+ *         cdef double total = 0.0             # <<<<<<<<<<<<<<
+ *         cdef double H_left
+ *         cdef double H_right
+ */
+  __pyx_v_total = 0.0;
+
+  /* "sklearn/tree/_tree.pyx":329
+ *         cdef double e1, e2
+ * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             H_left = 0.0
+ *             H_right = 0.0
  */
-  __pyx_t_1 = __pyx_v_self->__pyx_base.n_classes;
+  __pyx_t_1 = __pyx_v_n_outputs;
   for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
 
-    /* "sklearn/tree/_tree.pyx":233
+    /* "sklearn/tree/_tree.pyx":330
+ * 
+ *         for k from 0 <= k < n_outputs:
+ *             H_left = 0.0             # <<<<<<<<<<<<<<
+ *             H_right = 0.0
  * 
- *         for k from 0 <= k < self.n_classes:
- *             if self.label_count_left[k] > 0:             # <<<<<<<<<<<<<<
- *                 H_left -= ((self.label_count_left[k] / n_left)
- *                            * log(self.label_count_left[k] / n_left))
  */
-    __pyx_t_2 = ((__pyx_v_self->__pyx_base.label_count_left[__pyx_v_k]) > 0);
-    if (__pyx_t_2) {
+    __pyx_v_H_left = 0.0;
 
-      /* "sklearn/tree/_tree.pyx":235
- *             if self.label_count_left[k] > 0:
- *                 H_left -= ((self.label_count_left[k] / n_left)
- *                            * log(self.label_count_left[k] / n_left))             # <<<<<<<<<<<<<<
- *             if self.label_count_right[k] > 0:
- *                 H_right -= ((self.label_count_right[k] / n_right)
+    /* "sklearn/tree/_tree.pyx":331
+ *         for k from 0 <= k < n_outputs:
+ *             H_left = 0.0
+ *             H_right = 0.0             # <<<<<<<<<<<<<<
+ * 
+ *             for c from 0 <= c < n_classes[k]:
  */
-      __pyx_v_H_left = (__pyx_v_H_left - (((__pyx_v_self->__pyx_base.label_count_left[__pyx_v_k]) / __pyx_v_n_left) * log(((__pyx_v_self->__pyx_base.label_count_left[__pyx_v_k]) / __pyx_v_n_left))));
-      goto __pyx_L5;
-    }
-    __pyx_L5:;
+    __pyx_v_H_right = 0.0;
 
-    /* "sklearn/tree/_tree.pyx":236
- *                 H_left -= ((self.label_count_left[k] / n_left)
- *                            * log(self.label_count_left[k] / n_left))
- *             if self.label_count_right[k] > 0:             # <<<<<<<<<<<<<<
- *                 H_right -= ((self.label_count_right[k] / n_right)
- *                             * log(self.label_count_right[k] / n_right))
+    /* "sklearn/tree/_tree.pyx":333
+ *             H_right = 0.0
+ * 
+ *             for c from 0 <= c < n_classes[k]:             # <<<<<<<<<<<<<<
+ *                 if label_count_left[k * label_count_stride + c] > 0:
+ *                     H_left -= ((label_count_left[k * label_count_stride + c] / n_left) * log(label_count_left[k * label_count_stride + c] / n_left))
  */
-    __pyx_t_2 = ((__pyx_v_self->__pyx_base.label_count_right[__pyx_v_k]) > 0);
-    if (__pyx_t_2) {
+    __pyx_t_2 = (__pyx_v_n_classes[__pyx_v_k]);
+    for (__pyx_v_c = 0; __pyx_v_c < __pyx_t_2; __pyx_v_c++) {
 
-      /* "sklearn/tree/_tree.pyx":238
- *             if self.label_count_right[k] > 0:
- *                 H_right -= ((self.label_count_right[k] / n_right)
- *                             * log(self.label_count_right[k] / n_right))             # <<<<<<<<<<<<<<
+      /* "sklearn/tree/_tree.pyx":334
+ * 
+ *             for c from 0 <= c < n_classes[k]:
+ *                 if label_count_left[k * label_count_stride + c] > 0:             # <<<<<<<<<<<<<<
+ *                     H_left -= ((label_count_left[k * label_count_stride + c] / n_left) * log(label_count_left[k * label_count_stride + c] / n_left))
  * 
- *         e1 = (n_left / self.n_samples) * H_left
  */
-      __pyx_v_H_right = (__pyx_v_H_right - (((__pyx_v_self->__pyx_base.label_count_right[__pyx_v_k]) / __pyx_v_n_right) * log(((__pyx_v_self->__pyx_base.label_count_right[__pyx_v_k]) / __pyx_v_n_right))));
-      goto __pyx_L6;
+      __pyx_t_3 = ((__pyx_v_label_count_left[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]) > 0);
+      if (__pyx_t_3) {
+
+        /* "sklearn/tree/_tree.pyx":335
+ *             for c from 0 <= c < n_classes[k]:
+ *                 if label_count_left[k * label_count_stride + c] > 0:
+ *                     H_left -= ((label_count_left[k * label_count_stride + c] / n_left) * log(label_count_left[k * label_count_stride + c] / n_left))             # <<<<<<<<<<<<<<
+ * 
+ *                 if self.label_count_right[k * label_count_stride + c] > 0:
+ */
+        __pyx_v_H_left = (__pyx_v_H_left - (((__pyx_v_label_count_left[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]) / __pyx_v_n_left) * log(((__pyx_v_label_count_left[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]) / __pyx_v_n_left))));
+        goto __pyx_L7;
+      }
+      __pyx_L7:;
+
+      /* "sklearn/tree/_tree.pyx":337
+ *                     H_left -= ((label_count_left[k * label_count_stride + c] / n_left) * log(label_count_left[k * label_count_stride + c] / n_left))
+ * 
+ *                 if self.label_count_right[k * label_count_stride + c] > 0:             # <<<<<<<<<<<<<<
+ *                     H_right -= ((label_count_right[k * label_count_stride + c] / n_right) * log(label_count_right[k * label_count_stride + c] / n_right))
+ * 
+ */
+      __pyx_t_3 = ((__pyx_v_self->__pyx_base.label_count_right[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]) > 0);
+      if (__pyx_t_3) {
+
+        /* "sklearn/tree/_tree.pyx":338
+ * 
+ *                 if self.label_count_right[k * label_count_stride + c] > 0:
+ *                     H_right -= ((label_count_right[k * label_count_stride + c] / n_right) * log(label_count_right[k * label_count_stride + c] / n_right))             # <<<<<<<<<<<<<<
+ * 
+ *             e1 = (n_left / n_samples) * H_left
+ */
+        __pyx_v_H_right = (__pyx_v_H_right - (((__pyx_v_label_count_right[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]) / __pyx_v_n_right) * log(((__pyx_v_label_count_right[((__pyx_v_k * __pyx_v_label_count_stride) + __pyx_v_c)]) / __pyx_v_n_right))));
+        goto __pyx_L8;
+      }
+      __pyx_L8:;
     }
-    __pyx_L6:;
-  }
 
-  /* "sklearn/tree/_tree.pyx":240
- *                             * log(self.label_count_right[k] / n_right))
+    /* "sklearn/tree/_tree.pyx":340
+ *                     H_right -= ((label_count_right[k * label_count_stride + c] / n_right) * log(label_count_right[k * label_count_stride + c] / n_right))
+ * 
+ *             e1 = (n_left / n_samples) * H_left             # <<<<<<<<<<<<<<
+ *             e2 = (n_right / n_samples) * H_right
  * 
- *         e1 = (n_left / self.n_samples) * H_left             # <<<<<<<<<<<<<<
- *         e2 = (n_right / self.n_samples) * H_right
- *         return e1 + e2
  */
-  __pyx_v_e1 = ((__pyx_v_n_left / __pyx_v_self->__pyx_base.n_samples) * __pyx_v_H_left);
+    __pyx_v_e1 = ((__pyx_v_n_left / __pyx_v_n_samples) * __pyx_v_H_left);
 
-  /* "sklearn/tree/_tree.pyx":241
+    /* "sklearn/tree/_tree.pyx":341
  * 
- *         e1 = (n_left / self.n_samples) * H_left
- *         e2 = (n_right / self.n_samples) * H_right             # <<<<<<<<<<<<<<
- *         return e1 + e2
+ *             e1 = (n_left / n_samples) * H_left
+ *             e2 = (n_right / n_samples) * H_right             # <<<<<<<<<<<<<<
  * 
+ *             total += e1 + e2
  */
-  __pyx_v_e2 = ((__pyx_v_n_right / __pyx_v_self->__pyx_base.n_samples) * __pyx_v_H_right);
+    __pyx_v_e2 = ((__pyx_v_n_right / __pyx_v_n_samples) * __pyx_v_H_right);
 
-  /* "sklearn/tree/_tree.pyx":242
- *         e1 = (n_left / self.n_samples) * H_left
- *         e2 = (n_right / self.n_samples) * H_right
- *         return e1 + e2             # <<<<<<<<<<<<<<
+    /* "sklearn/tree/_tree.pyx":343
+ *             e2 = (n_right / n_samples) * H_right
  * 
+ *             total += e1 + e2             # <<<<<<<<<<<<<<
  * 
+ *         return total / n_outputs
  */
-  __pyx_r = (__pyx_v_e1 + __pyx_v_e2);
+    __pyx_v_total = (__pyx_v_total + (__pyx_v_e1 + __pyx_v_e2));
+  }
+
+  /* "sklearn/tree/_tree.pyx":345
+ *             total += e1 + e2
+ * 
+ *         return total / n_outputs             # <<<<<<<<<<<<<<
+ * 
+ * 
+ */
+  __pyx_r = (__pyx_v_total / __pyx_v_n_outputs);
   goto __pyx_L0;
 
   __pyx_r = 0;
@@ -2789,242 +3373,520 @@ static double __pyx_f_7sklearn_4tree_5_tree_7Entropy_eval(struct __pyx_obj_7skle
 /* Python wrapper */
 static int __pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_1__init__(PyObject *__pyx_v_self, PyObject *__pyx_args, PyObject *__pyx_kwds); /*proto*/
 static int __pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_1__init__(PyObject *__pyx_v_self, PyObject *__pyx_args, PyObject *__pyx_kwds) {
+  int __pyx_v_n_outputs;
+  static PyObject **__pyx_pyargnames[] = {&__pyx_n_s__n_outputs,0};
   int __pyx_r;
   __Pyx_RefNannyDeclarations
   __Pyx_RefNannySetupContext("__init__ (wrapper)", 0);
-  if (unlikely(PyTuple_GET_SIZE(__pyx_args) > 0)) {
-    __Pyx_RaiseArgtupleInvalid("__init__", 1, 0, 0, PyTuple_GET_SIZE(__pyx_args)); return -1;}
-  if (unlikely(__pyx_kwds) && unlikely(PyDict_Size(__pyx_kwds) > 0) && unlikely(!__Pyx_CheckKeywordStrings(__pyx_kwds, "__init__", 0))) return -1;
-  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion___init__(((struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *)__pyx_v_self));
+  {
+    PyObject* values[1] = {0};
+    if (unlikely(__pyx_kwds)) {
+      Py_ssize_t kw_args;
+      const Py_ssize_t pos_args = PyTuple_GET_SIZE(__pyx_args);
+      switch (pos_args) {
+        case  1: values[0] = PyTuple_GET_ITEM(__pyx_args, 0);
+        case  0: break;
+        default: goto __pyx_L5_argtuple_error;
+      }
+      kw_args = PyDict_Size(__pyx_kwds);
+      switch (pos_args) {
+        case  0:
+        values[0] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__n_outputs);
+        if (likely(values[0])) kw_args--;
+        else goto __pyx_L5_argtuple_error;
+      }
+      if (unlikely(kw_args > 0)) {
+        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "__init__") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 402; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+      }
+    } else if (PyTuple_GET_SIZE(__pyx_args) != 1) {
+      goto __pyx_L5_argtuple_error;
+    } else {
+      values[0] = PyTuple_GET_ITEM(__pyx_args, 0);
+    }
+    __pyx_v_n_outputs = __Pyx_PyInt_AsInt(values[0]); if (unlikely((__pyx_v_n_outputs == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 402; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+  }
+  goto __pyx_L4_argument_unpacking_done;
+  __pyx_L5_argtuple_error:;
+  __Pyx_RaiseArgtupleInvalid("__init__", 1, 1, 1, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 402; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+  __pyx_L3_error:;
+  __Pyx_AddTraceback("sklearn.tree._tree.RegressionCriterion.__init__", __pyx_clineno, __pyx_lineno, __pyx_filename);
+  __Pyx_RefNannyFinishContext();
+  return -1;
+  __pyx_L4_argument_unpacking_done:;
+  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion___init__(((struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *)__pyx_v_self), __pyx_v_n_outputs);
   __Pyx_RefNannyFinishContext();
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":299
- *     cdef double var_right
+/* "sklearn/tree/_tree.pyx":402
+ *     cdef int n_left
+ * 
+ *     def __init__(self, int n_outputs):             # <<<<<<<<<<<<<<
+ *         cdef int k = 0
  * 
- *     def __init__(self):             # <<<<<<<<<<<<<<
- *         self.n_samples = 0
- *         self.n_left = 0
  */
 
-static int __pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion___init__(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self) {
+static int __pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion___init__(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self, int __pyx_v_n_outputs) {
+  CYTHON_UNUSED int __pyx_v_k;
   int __pyx_r;
   __Pyx_RefNannyDeclarations
   __Pyx_RefNannySetupContext("__init__", 0);
 
-  /* "sklearn/tree/_tree.pyx":300
+  /* "sklearn/tree/_tree.pyx":403
+ * 
+ *     def __init__(self, int n_outputs):
+ *         cdef int k = 0             # <<<<<<<<<<<<<<
+ * 
+ *         self.n_outputs = n_outputs
+ */
+  __pyx_v_k = 0;
+
+  /* "sklearn/tree/_tree.pyx":405
+ *         cdef int k = 0
+ * 
+ *         self.n_outputs = n_outputs             # <<<<<<<<<<<<<<
+ * 
+ *         self.n_samples = 0
+ */
+  __pyx_v_self->n_outputs = __pyx_v_n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":407
+ *         self.n_outputs = n_outputs
  * 
- *     def __init__(self):
  *         self.n_samples = 0             # <<<<<<<<<<<<<<
  *         self.n_left = 0
  *         self.n_right = 0
  */
   __pyx_v_self->n_samples = 0;
 
-  /* "sklearn/tree/_tree.pyx":301
- *     def __init__(self):
+  /* "sklearn/tree/_tree.pyx":408
+ * 
  *         self.n_samples = 0
  *         self.n_left = 0             # <<<<<<<<<<<<<<
  *         self.n_right = 0
- *         self.mean_left = 0.0
+ * 
  */
   __pyx_v_self->n_left = 0;
 
-  /* "sklearn/tree/_tree.pyx":302
+  /* "sklearn/tree/_tree.pyx":409
  *         self.n_samples = 0
  *         self.n_left = 0
  *         self.n_right = 0             # <<<<<<<<<<<<<<
- *         self.mean_left = 0.0
- *         self.mean_right = 0.0
+ * 
+ *         self.mean_left = <double*> calloc(n_outputs, sizeof(double))
  */
   __pyx_v_self->n_right = 0;
 
-  /* "sklearn/tree/_tree.pyx":303
- *         self.n_left = 0
+  /* "sklearn/tree/_tree.pyx":411
  *         self.n_right = 0
- *         self.mean_left = 0.0             # <<<<<<<<<<<<<<
- *         self.mean_right = 0.0
- *         self.mean_init = 0.0
+ * 
+ *         self.mean_left = <double*> calloc(n_outputs, sizeof(double))             # <<<<<<<<<<<<<<
+ *         self.mean_right = <double*> calloc(n_outputs, sizeof(double))
+ *         self.mean_init = <double*> calloc(n_outputs, sizeof(double))
  */
-  __pyx_v_self->mean_left = 0.0;
+  __pyx_v_self->mean_left = ((double *)calloc(__pyx_v_n_outputs, (sizeof(double))));
 
-  /* "sklearn/tree/_tree.pyx":304
- *         self.n_right = 0
- *         self.mean_left = 0.0
- *         self.mean_right = 0.0             # <<<<<<<<<<<<<<
- *         self.mean_init = 0.0
- *         self.sq_sum_right = 0.0
+  /* "sklearn/tree/_tree.pyx":412
+ * 
+ *         self.mean_left = <double*> calloc(n_outputs, sizeof(double))
+ *         self.mean_right = <double*> calloc(n_outputs, sizeof(double))             # <<<<<<<<<<<<<<
+ *         self.mean_init = <double*> calloc(n_outputs, sizeof(double))
+ *         self.sq_sum_left = <double*> calloc(n_outputs, sizeof(double))
  */
-  __pyx_v_self->mean_right = 0.0;
+  __pyx_v_self->mean_right = ((double *)calloc(__pyx_v_n_outputs, (sizeof(double))));
 
-  /* "sklearn/tree/_tree.pyx":305
- *         self.mean_left = 0.0
- *         self.mean_right = 0.0
- *         self.mean_init = 0.0             # <<<<<<<<<<<<<<
- *         self.sq_sum_right = 0.0
- *         self.sq_sum_left = 0.0
+  /* "sklearn/tree/_tree.pyx":413
+ *         self.mean_left = <double*> calloc(n_outputs, sizeof(double))
+ *         self.mean_right = <double*> calloc(n_outputs, sizeof(double))
+ *         self.mean_init = <double*> calloc(n_outputs, sizeof(double))             # <<<<<<<<<<<<<<
+ *         self.sq_sum_left = <double*> calloc(n_outputs, sizeof(double))
+ *         self.sq_sum_right = <double*> calloc(n_outputs, sizeof(double))
  */
-  __pyx_v_self->mean_init = 0.0;
+  __pyx_v_self->mean_init = ((double *)calloc(__pyx_v_n_outputs, (sizeof(double))));
 
-  /* "sklearn/tree/_tree.pyx":306
- *         self.mean_right = 0.0
- *         self.mean_init = 0.0
- *         self.sq_sum_right = 0.0             # <<<<<<<<<<<<<<
- *         self.sq_sum_left = 0.0
- *         self.sq_sum_init = 0.0
+  /* "sklearn/tree/_tree.pyx":414
+ *         self.mean_right = <double*> calloc(n_outputs, sizeof(double))
+ *         self.mean_init = <double*> calloc(n_outputs, sizeof(double))
+ *         self.sq_sum_left = <double*> calloc(n_outputs, sizeof(double))             # <<<<<<<<<<<<<<
+ *         self.sq_sum_right = <double*> calloc(n_outputs, sizeof(double))
+ *         self.sq_sum_init = <double*> calloc(n_outputs, sizeof(double))
  */
-  __pyx_v_self->sq_sum_right = 0.0;
+  __pyx_v_self->sq_sum_left = ((double *)calloc(__pyx_v_n_outputs, (sizeof(double))));
 
-  /* "sklearn/tree/_tree.pyx":307
- *         self.mean_init = 0.0
- *         self.sq_sum_right = 0.0
- *         self.sq_sum_left = 0.0             # <<<<<<<<<<<<<<
- *         self.sq_sum_init = 0.0
- *         self.var_left = 0.0
+  /* "sklearn/tree/_tree.pyx":415
+ *         self.mean_init = <double*> calloc(n_outputs, sizeof(double))
+ *         self.sq_sum_left = <double*> calloc(n_outputs, sizeof(double))
+ *         self.sq_sum_right = <double*> calloc(n_outputs, sizeof(double))             # <<<<<<<<<<<<<<
+ *         self.sq_sum_init = <double*> calloc(n_outputs, sizeof(double))
+ *         self.var_left = <double*> calloc(n_outputs, sizeof(double))
  */
-  __pyx_v_self->sq_sum_left = 0.0;
+  __pyx_v_self->sq_sum_right = ((double *)calloc(__pyx_v_n_outputs, (sizeof(double))));
 
-  /* "sklearn/tree/_tree.pyx":308
- *         self.sq_sum_right = 0.0
- *         self.sq_sum_left = 0.0
- *         self.sq_sum_init = 0.0             # <<<<<<<<<<<<<<
- *         self.var_left = 0.0
- *         self.var_right = 0.0
+  /* "sklearn/tree/_tree.pyx":416
+ *         self.sq_sum_left = <double*> calloc(n_outputs, sizeof(double))
+ *         self.sq_sum_right = <double*> calloc(n_outputs, sizeof(double))
+ *         self.sq_sum_init = <double*> calloc(n_outputs, sizeof(double))             # <<<<<<<<<<<<<<
+ *         self.var_left = <double*> calloc(n_outputs, sizeof(double))
+ *         self.var_right = <double*> calloc(n_outputs, sizeof(double))
  */
-  __pyx_v_self->sq_sum_init = 0.0;
+  __pyx_v_self->sq_sum_init = ((double *)calloc(__pyx_v_n_outputs, (sizeof(double))));
 
-  /* "sklearn/tree/_tree.pyx":309
- *         self.sq_sum_left = 0.0
- *         self.sq_sum_init = 0.0
- *         self.var_left = 0.0             # <<<<<<<<<<<<<<
- *         self.var_right = 0.0
+  /* "sklearn/tree/_tree.pyx":417
+ *         self.sq_sum_right = <double*> calloc(n_outputs, sizeof(double))
+ *         self.sq_sum_init = <double*> calloc(n_outputs, sizeof(double))
+ *         self.var_left = <double*> calloc(n_outputs, sizeof(double))             # <<<<<<<<<<<<<<
+ *         self.var_right = <double*> calloc(n_outputs, sizeof(double))
  * 
  */
-  __pyx_v_self->var_left = 0.0;
+  __pyx_v_self->var_left = ((double *)calloc(__pyx_v_n_outputs, (sizeof(double))));
 
-  /* "sklearn/tree/_tree.pyx":310
- *         self.sq_sum_init = 0.0
- *         self.var_left = 0.0
- *         self.var_right = 0.0             # <<<<<<<<<<<<<<
+  /* "sklearn/tree/_tree.pyx":418
+ *         self.sq_sum_init = <double*> calloc(n_outputs, sizeof(double))
+ *         self.var_left = <double*> calloc(n_outputs, sizeof(double))
+ *         self.var_right = <double*> calloc(n_outputs, sizeof(double))             # <<<<<<<<<<<<<<
  * 
- *     cdef void init(self, DTYPE_t *y, BOOL_t *sample_mask, int n_samples,
+ *     def __del__(self):
  */
-  __pyx_v_self->var_right = 0.0;
+  __pyx_v_self->var_right = ((double *)calloc(__pyx_v_n_outputs, (sizeof(double))));
 
   __pyx_r = 0;
   __Pyx_RefNannyFinishContext();
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":312
- *         self.var_right = 0.0
+/* Python wrapper */
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_3__del__(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_3__del__(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused) {
+  PyObject *__pyx_r = 0;
+  __Pyx_RefNannyDeclarations
+  __Pyx_RefNannySetupContext("__del__ (wrapper)", 0);
+  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_2__del__(((struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *)__pyx_v_self));
+  __Pyx_RefNannyFinishContext();
+  return __pyx_r;
+}
+
+/* "sklearn/tree/_tree.pyx":420
+ *         self.var_right = <double*> calloc(n_outputs, sizeof(double))
+ * 
+ *     def __del__(self):             # <<<<<<<<<<<<<<
+ *         free(self.mean_left)
+ *         free(self.mean_right)
+ */
+
+static PyObject *__pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_2__del__(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self) {
+  PyObject *__pyx_r = NULL;
+  __Pyx_RefNannyDeclarations
+  __Pyx_RefNannySetupContext("__del__", 0);
+
+  /* "sklearn/tree/_tree.pyx":421
+ * 
+ *     def __del__(self):
+ *         free(self.mean_left)             # <<<<<<<<<<<<<<
+ *         free(self.mean_right)
+ *         free(self.mean_init)
+ */
+  free(__pyx_v_self->mean_left);
+
+  /* "sklearn/tree/_tree.pyx":422
+ *     def __del__(self):
+ *         free(self.mean_left)
+ *         free(self.mean_right)             # <<<<<<<<<<<<<<
+ *         free(self.mean_init)
+ *         free(self.sq_sum_left)
+ */
+  free(__pyx_v_self->mean_right);
+
+  /* "sklearn/tree/_tree.pyx":423
+ *         free(self.mean_left)
+ *         free(self.mean_right)
+ *         free(self.mean_init)             # <<<<<<<<<<<<<<
+ *         free(self.sq_sum_left)
+ *         free(self.sq_sum_right)
+ */
+  free(__pyx_v_self->mean_init);
+
+  /* "sklearn/tree/_tree.pyx":424
+ *         free(self.mean_right)
+ *         free(self.mean_init)
+ *         free(self.sq_sum_left)             # <<<<<<<<<<<<<<
+ *         free(self.sq_sum_right)
+ *         free(self.sq_sum_init)
+ */
+  free(__pyx_v_self->sq_sum_left);
+
+  /* "sklearn/tree/_tree.pyx":425
+ *         free(self.mean_init)
+ *         free(self.sq_sum_left)
+ *         free(self.sq_sum_right)             # <<<<<<<<<<<<<<
+ *         free(self.sq_sum_init)
+ *         free(self.var_left)
+ */
+  free(__pyx_v_self->sq_sum_right);
+
+  /* "sklearn/tree/_tree.pyx":426
+ *         free(self.sq_sum_left)
+ *         free(self.sq_sum_right)
+ *         free(self.sq_sum_init)             # <<<<<<<<<<<<<<
+ *         free(self.var_left)
+ *         free(self.var_right)
+ */
+  free(__pyx_v_self->sq_sum_init);
+
+  /* "sklearn/tree/_tree.pyx":427
+ *         free(self.sq_sum_right)
+ *         free(self.sq_sum_init)
+ *         free(self.var_left)             # <<<<<<<<<<<<<<
+ *         free(self.var_right)
+ * 
+ */
+  free(__pyx_v_self->var_left);
+
+  /* "sklearn/tree/_tree.pyx":428
+ *         free(self.sq_sum_init)
+ *         free(self.var_left)
+ *         free(self.var_right)             # <<<<<<<<<<<<<<
  * 
- *     cdef void init(self, DTYPE_t *y, BOOL_t *sample_mask, int n_samples,             # <<<<<<<<<<<<<<
- *                    int n_total_samples):
- *         """Initialise the criterion class; assume all samples
+ *     cdef void init(self, DTYPE_t* y,
  */
+  free(__pyx_v_self->var_right);
 
-static void __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask, int __pyx_v_n_samples, int __pyx_v_n_total_samples) {
+  __pyx_r = Py_None; __Pyx_INCREF(Py_None);
+  __Pyx_XGIVEREF(__pyx_r);
+  __Pyx_RefNannyFinishContext();
+  return __pyx_r;
+}
+
+/* "sklearn/tree/_tree.pyx":430
+ *         free(self.var_right)
+ * 
+ *     cdef void init(self, DTYPE_t* y,             # <<<<<<<<<<<<<<
+ *                          int y_stride,
+ *                          BOOL_t* sample_mask,
+ */
+
+static void __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, int __pyx_v_y_stride, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask, int __pyx_v_n_samples, int __pyx_v_n_total_samples) {
+  double *__pyx_v_mean_left;
+  double *__pyx_v_mean_right;
+  double *__pyx_v_mean_init;
+  double *__pyx_v_sq_sum_left;
+  double *__pyx_v_sq_sum_right;
+  double *__pyx_v_sq_sum_init;
+  double *__pyx_v_var_left;
+  double *__pyx_v_var_right;
+  int __pyx_v_n_outputs;
+  int __pyx_v_k;
   int __pyx_v_j;
+  __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_v_y_jk;
   __Pyx_RefNannyDeclarations
   int __pyx_t_1;
   int __pyx_t_2;
+  int __pyx_t_3;
+  int __pyx_t_4;
   __Pyx_RefNannySetupContext("init", 0);
 
-  /* "sklearn/tree/_tree.pyx":317
+  /* "sklearn/tree/_tree.pyx":438
  *            are in the right branch and store the mean and squared
  *            sum in `self.mean_init` and `self.sq_sum_init`. """
- *         self.mean_left = 0.0             # <<<<<<<<<<<<<<
- *         self.mean_right = 0.0
- *         self.mean_init = 0.0
+ *         cdef double* mean_left = self.mean_left             # <<<<<<<<<<<<<<
+ *         cdef double* mean_right = self.mean_right
+ *         cdef double* mean_init = self.mean_init
  */
-  __pyx_v_self->mean_left = 0.0;
+  __pyx_v_mean_left = __pyx_v_self->mean_left;
 
-  /* "sklearn/tree/_tree.pyx":318
+  /* "sklearn/tree/_tree.pyx":439
  *            sum in `self.mean_init` and `self.sq_sum_init`. """
- *         self.mean_left = 0.0
- *         self.mean_right = 0.0             # <<<<<<<<<<<<<<
- *         self.mean_init = 0.0
- *         self.sq_sum_right = 0.0
+ *         cdef double* mean_left = self.mean_left
+ *         cdef double* mean_right = self.mean_right             # <<<<<<<<<<<<<<
+ *         cdef double* mean_init = self.mean_init
+ *         cdef double* sq_sum_left = self.sq_sum_left
  */
-  __pyx_v_self->mean_right = 0.0;
+  __pyx_v_mean_right = __pyx_v_self->mean_right;
 
-  /* "sklearn/tree/_tree.pyx":319
- *         self.mean_left = 0.0
- *         self.mean_right = 0.0
- *         self.mean_init = 0.0             # <<<<<<<<<<<<<<
- *         self.sq_sum_right = 0.0
- *         self.sq_sum_left = 0.0
+  /* "sklearn/tree/_tree.pyx":440
+ *         cdef double* mean_left = self.mean_left
+ *         cdef double* mean_right = self.mean_right
+ *         cdef double* mean_init = self.mean_init             # <<<<<<<<<<<<<<
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ *         cdef double* sq_sum_right = self.sq_sum_right
  */
-  __pyx_v_self->mean_init = 0.0;
+  __pyx_v_mean_init = __pyx_v_self->mean_init;
 
-  /* "sklearn/tree/_tree.pyx":320
- *         self.mean_right = 0.0
- *         self.mean_init = 0.0
- *         self.sq_sum_right = 0.0             # <<<<<<<<<<<<<<
- *         self.sq_sum_left = 0.0
- *         self.sq_sum_init = 0.0
+  /* "sklearn/tree/_tree.pyx":441
+ *         cdef double* mean_right = self.mean_right
+ *         cdef double* mean_init = self.mean_init
+ *         cdef double* sq_sum_left = self.sq_sum_left             # <<<<<<<<<<<<<<
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ *         cdef double* sq_sum_init = self.sq_sum_init
  */
-  __pyx_v_self->sq_sum_right = 0.0;
+  __pyx_v_sq_sum_left = __pyx_v_self->sq_sum_left;
 
-  /* "sklearn/tree/_tree.pyx":321
- *         self.mean_init = 0.0
- *         self.sq_sum_right = 0.0
- *         self.sq_sum_left = 0.0             # <<<<<<<<<<<<<<
- *         self.sq_sum_init = 0.0
- *         self.var_left = 0.0
+  /* "sklearn/tree/_tree.pyx":442
+ *         cdef double* mean_init = self.mean_init
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ *         cdef double* sq_sum_right = self.sq_sum_right             # <<<<<<<<<<<<<<
+ *         cdef double* sq_sum_init = self.sq_sum_init
+ *         cdef double* var_left = self.var_left
  */
-  __pyx_v_self->sq_sum_left = 0.0;
+  __pyx_v_sq_sum_right = __pyx_v_self->sq_sum_right;
 
-  /* "sklearn/tree/_tree.pyx":322
- *         self.sq_sum_right = 0.0
- *         self.sq_sum_left = 0.0
- *         self.sq_sum_init = 0.0             # <<<<<<<<<<<<<<
- *         self.var_left = 0.0
- *         self.var_right = 0.0
+  /* "sklearn/tree/_tree.pyx":443
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ *         cdef double* sq_sum_init = self.sq_sum_init             # <<<<<<<<<<<<<<
+ *         cdef double* var_left = self.var_left
+ *         cdef double* var_right = self.var_right
  */
-  __pyx_v_self->sq_sum_init = 0.0;
+  __pyx_v_sq_sum_init = __pyx_v_self->sq_sum_init;
 
-  /* "sklearn/tree/_tree.pyx":323
- *         self.sq_sum_left = 0.0
- *         self.sq_sum_init = 0.0
- *         self.var_left = 0.0             # <<<<<<<<<<<<<<
- *         self.var_right = 0.0
- *         self.n_samples = n_samples
+  /* "sklearn/tree/_tree.pyx":444
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ *         cdef double* sq_sum_init = self.sq_sum_init
+ *         cdef double* var_left = self.var_left             # <<<<<<<<<<<<<<
+ *         cdef double* var_right = self.var_right
+ *         cdef int n_outputs = self.n_outputs
  */
-  __pyx_v_self->var_left = 0.0;
+  __pyx_v_var_left = __pyx_v_self->var_left;
 
-  /* "sklearn/tree/_tree.pyx":324
- *         self.sq_sum_init = 0.0
- *         self.var_left = 0.0
- *         self.var_right = 0.0             # <<<<<<<<<<<<<<
- *         self.n_samples = n_samples
+  /* "sklearn/tree/_tree.pyx":445
+ *         cdef double* sq_sum_init = self.sq_sum_init
+ *         cdef double* var_left = self.var_left
+ *         cdef double* var_right = self.var_right             # <<<<<<<<<<<<<<
+ *         cdef int n_outputs = self.n_outputs
+ * 
+ */
+  __pyx_v_var_right = __pyx_v_self->var_right;
+
+  /* "sklearn/tree/_tree.pyx":446
+ *         cdef double* var_left = self.var_left
+ *         cdef double* var_right = self.var_right
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int k = 0
+ */
+  __pyx_v_n_outputs = __pyx_v_self->n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":448
+ *         cdef int n_outputs = self.n_outputs
  * 
+ *         cdef int k = 0             # <<<<<<<<<<<<<<
+ * 
+ *         for k from 0 <= k < n_outputs:
+ */
+  __pyx_v_k = 0;
+
+  /* "sklearn/tree/_tree.pyx":450
+ *         cdef int k = 0
+ * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             mean_left[k] = 0.0
+ *             mean_right[k] = 0.0
+ */
+  __pyx_t_1 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
+
+    /* "sklearn/tree/_tree.pyx":451
+ * 
+ *         for k from 0 <= k < n_outputs:
+ *             mean_left[k] = 0.0             # <<<<<<<<<<<<<<
+ *             mean_right[k] = 0.0
+ *             mean_init[k] = 0.0
+ */
+    (__pyx_v_mean_left[__pyx_v_k]) = 0.0;
+
+    /* "sklearn/tree/_tree.pyx":452
+ *         for k from 0 <= k < n_outputs:
+ *             mean_left[k] = 0.0
+ *             mean_right[k] = 0.0             # <<<<<<<<<<<<<<
+ *             mean_init[k] = 0.0
+ *             sq_sum_right[k] = 0.0
+ */
+    (__pyx_v_mean_right[__pyx_v_k]) = 0.0;
+
+    /* "sklearn/tree/_tree.pyx":453
+ *             mean_left[k] = 0.0
+ *             mean_right[k] = 0.0
+ *             mean_init[k] = 0.0             # <<<<<<<<<<<<<<
+ *             sq_sum_right[k] = 0.0
+ *             sq_sum_left[k] = 0.0
+ */
+    (__pyx_v_mean_init[__pyx_v_k]) = 0.0;
+
+    /* "sklearn/tree/_tree.pyx":454
+ *             mean_right[k] = 0.0
+ *             mean_init[k] = 0.0
+ *             sq_sum_right[k] = 0.0             # <<<<<<<<<<<<<<
+ *             sq_sum_left[k] = 0.0
+ *             sq_sum_init[k] = 0.0
+ */
+    (__pyx_v_sq_sum_right[__pyx_v_k]) = 0.0;
+
+    /* "sklearn/tree/_tree.pyx":455
+ *             mean_init[k] = 0.0
+ *             sq_sum_right[k] = 0.0
+ *             sq_sum_left[k] = 0.0             # <<<<<<<<<<<<<<
+ *             sq_sum_init[k] = 0.0
+ *             var_left[k] = 0.0
  */
-  __pyx_v_self->var_right = 0.0;
+    (__pyx_v_sq_sum_left[__pyx_v_k]) = 0.0;
 
-  /* "sklearn/tree/_tree.pyx":325
- *         self.var_left = 0.0
- *         self.var_right = 0.0
+    /* "sklearn/tree/_tree.pyx":456
+ *             sq_sum_right[k] = 0.0
+ *             sq_sum_left[k] = 0.0
+ *             sq_sum_init[k] = 0.0             # <<<<<<<<<<<<<<
+ *             var_left[k] = 0.0
+ *             var_right[k] = 0.0
+ */
+    (__pyx_v_sq_sum_init[__pyx_v_k]) = 0.0;
+
+    /* "sklearn/tree/_tree.pyx":457
+ *             sq_sum_left[k] = 0.0
+ *             sq_sum_init[k] = 0.0
+ *             var_left[k] = 0.0             # <<<<<<<<<<<<<<
+ *             var_right[k] = 0.0
+ * 
+ */
+    (__pyx_v_var_left[__pyx_v_k]) = 0.0;
+
+    /* "sklearn/tree/_tree.pyx":458
+ *             sq_sum_init[k] = 0.0
+ *             var_left[k] = 0.0
+ *             var_right[k] = 0.0             # <<<<<<<<<<<<<<
+ * 
+ *         self.n_samples = n_samples
+ */
+    (__pyx_v_var_right[__pyx_v_k]) = 0.0;
+  }
+
+  /* "sklearn/tree/_tree.pyx":460
+ *             var_right[k] = 0.0
+ * 
  *         self.n_samples = n_samples             # <<<<<<<<<<<<<<
  * 
  *         cdef int j = 0
  */
   __pyx_v_self->n_samples = __pyx_v_n_samples;
 
-  /* "sklearn/tree/_tree.pyx":327
+  /* "sklearn/tree/_tree.pyx":462
  *         self.n_samples = n_samples
  * 
  *         cdef int j = 0             # <<<<<<<<<<<<<<
- *         for j from 0 <= j < n_total_samples:
- *             if sample_mask[j] == 0:
+ *         cdef DTYPE_t y_jk = 0.0
+ * 
  */
   __pyx_v_j = 0;
 
-  /* "sklearn/tree/_tree.pyx":328
+  /* "sklearn/tree/_tree.pyx":463
  * 
  *         cdef int j = 0
+ *         cdef DTYPE_t y_jk = 0.0             # <<<<<<<<<<<<<<
+ * 
+ *         for j from 0 <= j < n_total_samples:
+ */
+  __pyx_v_y_jk = 0.0;
+
+  /* "sklearn/tree/_tree.pyx":465
+ *         cdef DTYPE_t y_jk = 0.0
+ * 
  *         for j from 0 <= j < n_total_samples:             # <<<<<<<<<<<<<<
  *             if sample_mask[j] == 0:
  *                 continue
@@ -3032,59 +3894,93 @@ static void __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init(struct __py
   __pyx_t_1 = __pyx_v_n_total_samples;
   for (__pyx_v_j = 0; __pyx_v_j < __pyx_t_1; __pyx_v_j++) {
 
-    /* "sklearn/tree/_tree.pyx":329
- *         cdef int j = 0
+    /* "sklearn/tree/_tree.pyx":466
+ * 
  *         for j from 0 <= j < n_total_samples:
  *             if sample_mask[j] == 0:             # <<<<<<<<<<<<<<
  *                 continue
- *             self.sq_sum_init += (y[j] * y[j])
+ * 
  */
     __pyx_t_2 = ((__pyx_v_sample_mask[__pyx_v_j]) == 0);
     if (__pyx_t_2) {
 
-      /* "sklearn/tree/_tree.pyx":330
+      /* "sklearn/tree/_tree.pyx":467
  *         for j from 0 <= j < n_total_samples:
  *             if sample_mask[j] == 0:
  *                 continue             # <<<<<<<<<<<<<<
- *             self.sq_sum_init += (y[j] * y[j])
- *             self.mean_init += y[j]
+ * 
+ *             for k from 0 <= k < n_outputs:
  */
-      goto __pyx_L3_continue;
-      goto __pyx_L5;
+      goto __pyx_L5_continue;
+      goto __pyx_L7;
     }
-    __pyx_L5:;
+    __pyx_L7:;
 
-    /* "sklearn/tree/_tree.pyx":331
- *             if sample_mask[j] == 0:
+    /* "sklearn/tree/_tree.pyx":469
  *                 continue
- *             self.sq_sum_init += (y[j] * y[j])             # <<<<<<<<<<<<<<
- *             self.mean_init += y[j]
  * 
+ *             for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *                 y_jk = y[j * y_stride + k]
+ *                 sq_sum_init[k] += y_jk * y_jk
  */
-    __pyx_v_self->sq_sum_init = (__pyx_v_self->sq_sum_init + ((__pyx_v_y[__pyx_v_j]) * (__pyx_v_y[__pyx_v_j])));
+    __pyx_t_3 = __pyx_v_n_outputs;
+    for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_3; __pyx_v_k++) {
 
-    /* "sklearn/tree/_tree.pyx":332
- *                 continue
- *             self.sq_sum_init += (y[j] * y[j])
- *             self.mean_init += y[j]             # <<<<<<<<<<<<<<
+      /* "sklearn/tree/_tree.pyx":470
  * 
- *         self.mean_init = self.mean_init / self.n_samples
+ *             for k from 0 <= k < n_outputs:
+ *                 y_jk = y[j * y_stride + k]             # <<<<<<<<<<<<<<
+ *                 sq_sum_init[k] += y_jk * y_jk
+ *                 mean_init[k] += y_jk
  */
-    __pyx_v_self->mean_init = (__pyx_v_self->mean_init + (__pyx_v_y[__pyx_v_j]));
-    __pyx_L3_continue:;
+      __pyx_v_y_jk = (__pyx_v_y[((__pyx_v_j * __pyx_v_y_stride) + __pyx_v_k)]);
+
+      /* "sklearn/tree/_tree.pyx":471
+ *             for k from 0 <= k < n_outputs:
+ *                 y_jk = y[j * y_stride + k]
+ *                 sq_sum_init[k] += y_jk * y_jk             # <<<<<<<<<<<<<<
+ *                 mean_init[k] += y_jk
+ * 
+ */
+      __pyx_t_4 = __pyx_v_k;
+      (__pyx_v_sq_sum_init[__pyx_t_4]) = ((__pyx_v_sq_sum_init[__pyx_t_4]) + (__pyx_v_y_jk * __pyx_v_y_jk));
+
+      /* "sklearn/tree/_tree.pyx":472
+ *                 y_jk = y[j * y_stride + k]
+ *                 sq_sum_init[k] += y_jk * y_jk
+ *                 mean_init[k] += y_jk             # <<<<<<<<<<<<<<
+ * 
+ *         for k from 0 <= k < n_outputs:
+ */
+      __pyx_t_4 = __pyx_v_k;
+      (__pyx_v_mean_init[__pyx_t_4]) = ((__pyx_v_mean_init[__pyx_t_4]) + __pyx_v_y_jk);
+    }
+    __pyx_L5_continue:;
   }
 
-  /* "sklearn/tree/_tree.pyx":334
- *             self.mean_init += y[j]
+  /* "sklearn/tree/_tree.pyx":474
+ *                 mean_init[k] += y_jk
+ * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             mean_init[k] /= n_samples
+ * 
+ */
+  __pyx_t_1 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
+
+    /* "sklearn/tree/_tree.pyx":475
  * 
- *         self.mean_init = self.mean_init / self.n_samples             # <<<<<<<<<<<<<<
+ *         for k from 0 <= k < n_outputs:
+ *             mean_init[k] /= n_samples             # <<<<<<<<<<<<<<
  * 
  *         self.reset()
  */
-  __pyx_v_self->mean_init = (__pyx_v_self->mean_init / __pyx_v_self->n_samples);
+    __pyx_t_3 = __pyx_v_k;
+    (__pyx_v_mean_init[__pyx_t_3]) = ((__pyx_v_mean_init[__pyx_t_3]) / __pyx_v_n_samples);
+  }
 
-  /* "sklearn/tree/_tree.pyx":336
- *         self.mean_init = self.mean_init / self.n_samples
+  /* "sklearn/tree/_tree.pyx":477
+ *             mean_init[k] /= n_samples
  * 
  *         self.reset()             # <<<<<<<<<<<<<<
  * 
@@ -3095,7 +3991,7 @@ static void __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init(struct __py
   __Pyx_RefNannyFinishContext();
 }
 
-/* "sklearn/tree/_tree.pyx":338
+/* "sklearn/tree/_tree.pyx":479
  *         self.reset()
  * 
  *     cdef void reset(self):             # <<<<<<<<<<<<<<
@@ -3104,243 +4000,510 @@ static void __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init(struct __py
  */
 
 static void __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_reset(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self) {
+  double *__pyx_v_mean_left;
+  double *__pyx_v_mean_right;
+  double *__pyx_v_mean_init;
+  double *__pyx_v_sq_sum_left;
+  double *__pyx_v_sq_sum_right;
+  double *__pyx_v_sq_sum_init;
+  double *__pyx_v_var_left;
+  double *__pyx_v_var_right;
+  int __pyx_v_n_samples;
+  int __pyx_v_n_outputs;
+  int __pyx_v_k;
   __Pyx_RefNannyDeclarations
+  int __pyx_t_1;
   __Pyx_RefNannySetupContext("reset", 0);
 
-  /* "sklearn/tree/_tree.pyx":345
+  /* "sklearn/tree/_tree.pyx":486
  *         right branch.
  *         """
+ *         cdef double* mean_left = self.mean_left             # <<<<<<<<<<<<<<
+ *         cdef double* mean_right = self.mean_right
+ *         cdef double* mean_init = self.mean_init
+ */
+  __pyx_v_mean_left = __pyx_v_self->mean_left;
+
+  /* "sklearn/tree/_tree.pyx":487
+ *         """
+ *         cdef double* mean_left = self.mean_left
+ *         cdef double* mean_right = self.mean_right             # <<<<<<<<<<<<<<
+ *         cdef double* mean_init = self.mean_init
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ */
+  __pyx_v_mean_right = __pyx_v_self->mean_right;
+
+  /* "sklearn/tree/_tree.pyx":488
+ *         cdef double* mean_left = self.mean_left
+ *         cdef double* mean_right = self.mean_right
+ *         cdef double* mean_init = self.mean_init             # <<<<<<<<<<<<<<
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ */
+  __pyx_v_mean_init = __pyx_v_self->mean_init;
+
+  /* "sklearn/tree/_tree.pyx":489
+ *         cdef double* mean_right = self.mean_right
+ *         cdef double* mean_init = self.mean_init
+ *         cdef double* sq_sum_left = self.sq_sum_left             # <<<<<<<<<<<<<<
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ *         cdef double* sq_sum_init = self.sq_sum_init
+ */
+  __pyx_v_sq_sum_left = __pyx_v_self->sq_sum_left;
+
+  /* "sklearn/tree/_tree.pyx":490
+ *         cdef double* mean_init = self.mean_init
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ *         cdef double* sq_sum_right = self.sq_sum_right             # <<<<<<<<<<<<<<
+ *         cdef double* sq_sum_init = self.sq_sum_init
+ *         cdef double* var_left = self.var_left
+ */
+  __pyx_v_sq_sum_right = __pyx_v_self->sq_sum_right;
+
+  /* "sklearn/tree/_tree.pyx":491
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ *         cdef double* sq_sum_init = self.sq_sum_init             # <<<<<<<<<<<<<<
+ *         cdef double* var_left = self.var_left
+ *         cdef double* var_right = self.var_right
+ */
+  __pyx_v_sq_sum_init = __pyx_v_self->sq_sum_init;
+
+  /* "sklearn/tree/_tree.pyx":492
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ *         cdef double* sq_sum_init = self.sq_sum_init
+ *         cdef double* var_left = self.var_left             # <<<<<<<<<<<<<<
+ *         cdef double* var_right = self.var_right
+ * 
+ */
+  __pyx_v_var_left = __pyx_v_self->var_left;
+
+  /* "sklearn/tree/_tree.pyx":493
+ *         cdef double* sq_sum_init = self.sq_sum_init
+ *         cdef double* var_left = self.var_left
+ *         cdef double* var_right = self.var_right             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int n_samples = self.n_samples
+ */
+  __pyx_v_var_right = __pyx_v_self->var_right;
+
+  /* "sklearn/tree/_tree.pyx":495
+ *         cdef double* var_right = self.var_right
+ * 
+ *         cdef int n_samples = self.n_samples             # <<<<<<<<<<<<<<
+ *         cdef int n_outputs = self.n_outputs
+ * 
+ */
+  __pyx_v_n_samples = __pyx_v_self->n_samples;
+
+  /* "sklearn/tree/_tree.pyx":496
+ * 
+ *         cdef int n_samples = self.n_samples
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int k = 0
+ */
+  __pyx_v_n_outputs = __pyx_v_self->n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":498
+ *         cdef int n_outputs = self.n_outputs
+ * 
+ *         cdef int k = 0             # <<<<<<<<<<<<<<
+ * 
+ *         self.n_right = self.n_samples
+ */
+  __pyx_v_k = 0;
+
+  /* "sklearn/tree/_tree.pyx":500
+ *         cdef int k = 0
+ * 
  *         self.n_right = self.n_samples             # <<<<<<<<<<<<<<
  *         self.n_left = 0
- *         self.mean_right = self.mean_init
+ * 
  */
   __pyx_v_self->n_right = __pyx_v_self->n_samples;
 
-  /* "sklearn/tree/_tree.pyx":346
- *         """
+  /* "sklearn/tree/_tree.pyx":501
+ * 
  *         self.n_right = self.n_samples
  *         self.n_left = 0             # <<<<<<<<<<<<<<
- *         self.mean_right = self.mean_init
- *         self.mean_left = 0.0
+ * 
+ *         for k from 0 <= k < n_outputs:
  */
   __pyx_v_self->n_left = 0;
 
-  /* "sklearn/tree/_tree.pyx":347
- *         self.n_right = self.n_samples
+  /* "sklearn/tree/_tree.pyx":503
  *         self.n_left = 0
- *         self.mean_right = self.mean_init             # <<<<<<<<<<<<<<
- *         self.mean_left = 0.0
- *         self.sq_sum_right = self.sq_sum_init
+ * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             mean_right[k] = mean_init[k]
+ *             mean_left[k] = 0.0
+ */
+  __pyx_t_1 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
+
+    /* "sklearn/tree/_tree.pyx":504
+ * 
+ *         for k from 0 <= k < n_outputs:
+ *             mean_right[k] = mean_init[k]             # <<<<<<<<<<<<<<
+ *             mean_left[k] = 0.0
+ *             sq_sum_right[k] = sq_sum_init[k]
  */
-  __pyx_v_self->mean_right = __pyx_v_self->mean_init;
+    (__pyx_v_mean_right[__pyx_v_k]) = (__pyx_v_mean_init[__pyx_v_k]);
 
-  /* "sklearn/tree/_tree.pyx":348
- *         self.n_left = 0
- *         self.mean_right = self.mean_init
- *         self.mean_left = 0.0             # <<<<<<<<<<<<<<
- *         self.sq_sum_right = self.sq_sum_init
- *         self.sq_sum_left = 0.0
+    /* "sklearn/tree/_tree.pyx":505
+ *         for k from 0 <= k < n_outputs:
+ *             mean_right[k] = mean_init[k]
+ *             mean_left[k] = 0.0             # <<<<<<<<<<<<<<
+ *             sq_sum_right[k] = sq_sum_init[k]
+ *             sq_sum_left[k] = 0.0
  */
-  __pyx_v_self->mean_left = 0.0;
+    (__pyx_v_mean_left[__pyx_v_k]) = 0.0;
 
-  /* "sklearn/tree/_tree.pyx":349
- *         self.mean_right = self.mean_init
- *         self.mean_left = 0.0
- *         self.sq_sum_right = self.sq_sum_init             # <<<<<<<<<<<<<<
- *         self.sq_sum_left = 0.0
- *         self.var_left = 0.0
+    /* "sklearn/tree/_tree.pyx":506
+ *             mean_right[k] = mean_init[k]
+ *             mean_left[k] = 0.0
+ *             sq_sum_right[k] = sq_sum_init[k]             # <<<<<<<<<<<<<<
+ *             sq_sum_left[k] = 0.0
+ *             var_left[k] = 0.0
  */
-  __pyx_v_self->sq_sum_right = __pyx_v_self->sq_sum_init;
+    (__pyx_v_sq_sum_right[__pyx_v_k]) = (__pyx_v_sq_sum_init[__pyx_v_k]);
 
-  /* "sklearn/tree/_tree.pyx":350
- *         self.mean_left = 0.0
- *         self.sq_sum_right = self.sq_sum_init
- *         self.sq_sum_left = 0.0             # <<<<<<<<<<<<<<
- *         self.var_left = 0.0
- *         self.var_right = self.sq_sum_right - \
+    /* "sklearn/tree/_tree.pyx":507
+ *             mean_left[k] = 0.0
+ *             sq_sum_right[k] = sq_sum_init[k]
+ *             sq_sum_left[k] = 0.0             # <<<<<<<<<<<<<<
+ *             var_left[k] = 0.0
+ *             var_right[k] = sq_sum_right[k] - n_samples * (mean_right[k] * mean_right[k])
  */
-  __pyx_v_self->sq_sum_left = 0.0;
+    (__pyx_v_sq_sum_left[__pyx_v_k]) = 0.0;
 
-  /* "sklearn/tree/_tree.pyx":351
- *         self.sq_sum_right = self.sq_sum_init
- *         self.sq_sum_left = 0.0
- *         self.var_left = 0.0             # <<<<<<<<<<<<<<
- *         self.var_right = self.sq_sum_right - \
- *             self.n_samples * (self.mean_right * self.mean_right)
+    /* "sklearn/tree/_tree.pyx":508
+ *             sq_sum_right[k] = sq_sum_init[k]
+ *             sq_sum_left[k] = 0.0
+ *             var_left[k] = 0.0             # <<<<<<<<<<<<<<
+ *             var_right[k] = sq_sum_right[k] - n_samples * (mean_right[k] * mean_right[k])
+ * 
  */
-  __pyx_v_self->var_left = 0.0;
+    (__pyx_v_var_left[__pyx_v_k]) = 0.0;
 
-  /* "sklearn/tree/_tree.pyx":352
- *         self.sq_sum_left = 0.0
- *         self.var_left = 0.0
- *         self.var_right = self.sq_sum_right - \             # <<<<<<<<<<<<<<
- *             self.n_samples * (self.mean_right * self.mean_right)
+    /* "sklearn/tree/_tree.pyx":509
+ *             sq_sum_left[k] = 0.0
+ *             var_left[k] = 0.0
+ *             var_right[k] = sq_sum_right[k] - n_samples * (mean_right[k] * mean_right[k])             # <<<<<<<<<<<<<<
  * 
+ *     cdef int update(self, int a,
  */
-  __pyx_v_self->var_right = (__pyx_v_self->sq_sum_right - (__pyx_v_self->n_samples * (__pyx_v_self->mean_right * __pyx_v_self->mean_right)));
+    (__pyx_v_var_right[__pyx_v_k]) = ((__pyx_v_sq_sum_right[__pyx_v_k]) - (__pyx_v_n_samples * ((__pyx_v_mean_right[__pyx_v_k]) * (__pyx_v_mean_right[__pyx_v_k]))));
+  }
 
   __Pyx_RefNannyFinishContext();
 }
 
-/* "sklearn/tree/_tree.pyx":355
- *             self.n_samples * (self.mean_right * self.mean_right)
+/* "sklearn/tree/_tree.pyx":511
+ *             var_right[k] = sq_sum_right[k] - n_samples * (mean_right[k] * mean_right[k])
  * 
- *     cdef int update(self, int a, int b, DTYPE_t *y, int *X_argsorted_i,             # <<<<<<<<<<<<<<
- *                     BOOL_t *sample_mask):
- *         """Update the criteria for each value in interval [a,b) (where a and b
+ *     cdef int update(self, int a,             # <<<<<<<<<<<<<<
+ *                           int b,
+ *                           DTYPE_t* y,
  */
 
-static int __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_update(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self, int __pyx_v_a, int __pyx_v_b, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, int *__pyx_v_X_argsorted_i, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask) {
+static int __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_update(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self, int __pyx_v_a, int __pyx_v_b, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y, int __pyx_v_y_stride, int *__pyx_v_X_argsorted_i, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask) {
+  double *__pyx_v_mean_left;
+  double *__pyx_v_mean_right;
+  double *__pyx_v_sq_sum_left;
+  double *__pyx_v_sq_sum_right;
+  double *__pyx_v_var_left;
+  double *__pyx_v_var_right;
+  int __pyx_v_n_samples;
+  int __pyx_v_n_outputs;
+  int __pyx_v_n_left;
+  int __pyx_v_n_right;
   double __pyx_v_y_idx;
   int __pyx_v_idx;
   int __pyx_v_j;
+  int __pyx_v_k;
   int __pyx_r;
   __Pyx_RefNannyDeclarations
   int __pyx_t_1;
   int __pyx_t_2;
+  int __pyx_t_3;
+  int __pyx_t_4;
   __Pyx_RefNannySetupContext("update", 0);
 
-  /* "sklearn/tree/_tree.pyx":359
+  /* "sklearn/tree/_tree.pyx":519
  *         """Update the criteria for each value in interval [a,b) (where a and b
  *            are indices in `X_argsorted_i`)."""
+ *         cdef double* mean_left = self.mean_left             # <<<<<<<<<<<<<<
+ *         cdef double* mean_right = self.mean_right
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ */
+  __pyx_v_mean_left = __pyx_v_self->mean_left;
+
+  /* "sklearn/tree/_tree.pyx":520
+ *            are indices in `X_argsorted_i`)."""
+ *         cdef double* mean_left = self.mean_left
+ *         cdef double* mean_right = self.mean_right             # <<<<<<<<<<<<<<
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ */
+  __pyx_v_mean_right = __pyx_v_self->mean_right;
+
+  /* "sklearn/tree/_tree.pyx":521
+ *         cdef double* mean_left = self.mean_left
+ *         cdef double* mean_right = self.mean_right
+ *         cdef double* sq_sum_left = self.sq_sum_left             # <<<<<<<<<<<<<<
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ *         cdef double* var_left = self.var_left
+ */
+  __pyx_v_sq_sum_left = __pyx_v_self->sq_sum_left;
+
+  /* "sklearn/tree/_tree.pyx":522
+ *         cdef double* mean_right = self.mean_right
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ *         cdef double* sq_sum_right = self.sq_sum_right             # <<<<<<<<<<<<<<
+ *         cdef double* var_left = self.var_left
+ *         cdef double* var_right = self.var_right
+ */
+  __pyx_v_sq_sum_right = __pyx_v_self->sq_sum_right;
+
+  /* "sklearn/tree/_tree.pyx":523
+ *         cdef double* sq_sum_left = self.sq_sum_left
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ *         cdef double* var_left = self.var_left             # <<<<<<<<<<<<<<
+ *         cdef double* var_right = self.var_right
+ * 
+ */
+  __pyx_v_var_left = __pyx_v_self->var_left;
+
+  /* "sklearn/tree/_tree.pyx":524
+ *         cdef double* sq_sum_right = self.sq_sum_right
+ *         cdef double* var_left = self.var_left
+ *         cdef double* var_right = self.var_right             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int n_samples = self.n_samples
+ */
+  __pyx_v_var_right = __pyx_v_self->var_right;
+
+  /* "sklearn/tree/_tree.pyx":526
+ *         cdef double* var_right = self.var_right
+ * 
+ *         cdef int n_samples = self.n_samples             # <<<<<<<<<<<<<<
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int n_left = self.n_left
+ */
+  __pyx_v_n_samples = __pyx_v_self->n_samples;
+
+  /* "sklearn/tree/_tree.pyx":527
+ * 
+ *         cdef int n_samples = self.n_samples
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ *         cdef int n_left = self.n_left
+ *         cdef int n_right = self.n_right
+ */
+  __pyx_v_n_outputs = __pyx_v_self->n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":528
+ *         cdef int n_samples = self.n_samples
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int n_left = self.n_left             # <<<<<<<<<<<<<<
+ *         cdef int n_right = self.n_right
+ * 
+ */
+  __pyx_v_n_left = __pyx_v_self->n_left;
+
+  /* "sklearn/tree/_tree.pyx":529
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef int n_left = self.n_left
+ *         cdef int n_right = self.n_right             # <<<<<<<<<<<<<<
+ * 
+ *         cdef double y_idx = 0.0
+ */
+  __pyx_v_n_right = __pyx_v_self->n_right;
+
+  /* "sklearn/tree/_tree.pyx":531
+ *         cdef int n_right = self.n_right
+ * 
  *         cdef double y_idx = 0.0             # <<<<<<<<<<<<<<
- *         cdef int idx, j
- *         # post condition: all samples from [0:b) are on the left side
+ *         cdef int idx, j, k
+ * 
  */
   __pyx_v_y_idx = 0.0;
 
-  /* "sklearn/tree/_tree.pyx":362
- *         cdef int idx, j
+  /* "sklearn/tree/_tree.pyx":535
+ * 
  *         # post condition: all samples from [0:b) are on the left side
  *         for idx from a <= idx < b:             # <<<<<<<<<<<<<<
  *             j = X_argsorted_i[idx]
- *             if sample_mask[j] == 0:
+ * 
  */
   __pyx_t_1 = __pyx_v_b;
   for (__pyx_v_idx = __pyx_v_a; __pyx_v_idx < __pyx_t_1; __pyx_v_idx++) {
 
-    /* "sklearn/tree/_tree.pyx":363
+    /* "sklearn/tree/_tree.pyx":536
  *         # post condition: all samples from [0:b) are on the left side
  *         for idx from a <= idx < b:
  *             j = X_argsorted_i[idx]             # <<<<<<<<<<<<<<
+ * 
  *             if sample_mask[j] == 0:
- *                 continue
  */
     __pyx_v_j = (__pyx_v_X_argsorted_i[__pyx_v_idx]);
 
-    /* "sklearn/tree/_tree.pyx":364
- *         for idx from a <= idx < b:
+    /* "sklearn/tree/_tree.pyx":538
  *             j = X_argsorted_i[idx]
+ * 
  *             if sample_mask[j] == 0:             # <<<<<<<<<<<<<<
  *                 continue
- *             y_idx = y[j]
+ * 
  */
     __pyx_t_2 = ((__pyx_v_sample_mask[__pyx_v_j]) == 0);
     if (__pyx_t_2) {
 
-      /* "sklearn/tree/_tree.pyx":365
- *             j = X_argsorted_i[idx]
+      /* "sklearn/tree/_tree.pyx":539
+ * 
  *             if sample_mask[j] == 0:
  *                 continue             # <<<<<<<<<<<<<<
- *             y_idx = y[j]
- *             self.sq_sum_left = self.sq_sum_left + (y_idx * y_idx)
+ * 
+ *             for k from 0 <= k < n_outputs:
  */
       goto __pyx_L3_continue;
       goto __pyx_L5;
     }
     __pyx_L5:;
 
-    /* "sklearn/tree/_tree.pyx":366
- *             if sample_mask[j] == 0:
+    /* "sklearn/tree/_tree.pyx":541
  *                 continue
- *             y_idx = y[j]             # <<<<<<<<<<<<<<
- *             self.sq_sum_left = self.sq_sum_left + (y_idx * y_idx)
- *             self.sq_sum_right = self.sq_sum_right - (y_idx * y_idx)
+ * 
+ *             for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *                 y_idx = y[j * y_stride + k]
+ *                 sq_sum_left[k] += (y_idx * y_idx)
  */
-    __pyx_v_y_idx = (__pyx_v_y[__pyx_v_j]);
+    __pyx_t_3 = __pyx_v_n_outputs;
+    for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_3; __pyx_v_k++) {
 
-    /* "sklearn/tree/_tree.pyx":367
- *                 continue
- *             y_idx = y[j]
- *             self.sq_sum_left = self.sq_sum_left + (y_idx * y_idx)             # <<<<<<<<<<<<<<
- *             self.sq_sum_right = self.sq_sum_right - (y_idx * y_idx)
+      /* "sklearn/tree/_tree.pyx":542
  * 
+ *             for k from 0 <= k < n_outputs:
+ *                 y_idx = y[j * y_stride + k]             # <<<<<<<<<<<<<<
+ *                 sq_sum_left[k] += (y_idx * y_idx)
+ *                 sq_sum_right[k] -= (y_idx * y_idx)
  */
-    __pyx_v_self->sq_sum_left = (__pyx_v_self->sq_sum_left + (__pyx_v_y_idx * __pyx_v_y_idx));
+      __pyx_v_y_idx = (__pyx_v_y[((__pyx_v_j * __pyx_v_y_stride) + __pyx_v_k)]);
 
-    /* "sklearn/tree/_tree.pyx":368
- *             y_idx = y[j]
- *             self.sq_sum_left = self.sq_sum_left + (y_idx * y_idx)
- *             self.sq_sum_right = self.sq_sum_right - (y_idx * y_idx)             # <<<<<<<<<<<<<<
+      /* "sklearn/tree/_tree.pyx":543
+ *             for k from 0 <= k < n_outputs:
+ *                 y_idx = y[j * y_stride + k]
+ *                 sq_sum_left[k] += (y_idx * y_idx)             # <<<<<<<<<<<<<<
+ *                 sq_sum_right[k] -= (y_idx * y_idx)
  * 
- *             self.mean_left = (self.n_left * self.mean_left + y_idx) / \
  */
-    __pyx_v_self->sq_sum_right = (__pyx_v_self->sq_sum_right - (__pyx_v_y_idx * __pyx_v_y_idx));
+      __pyx_t_4 = __pyx_v_k;
+      (__pyx_v_sq_sum_left[__pyx_t_4]) = ((__pyx_v_sq_sum_left[__pyx_t_4]) + (__pyx_v_y_idx * __pyx_v_y_idx));
 
-    /* "sklearn/tree/_tree.pyx":370
- *             self.sq_sum_right = self.sq_sum_right - (y_idx * y_idx)
+      /* "sklearn/tree/_tree.pyx":544
+ *                 y_idx = y[j * y_stride + k]
+ *                 sq_sum_left[k] += (y_idx * y_idx)
+ *                 sq_sum_right[k] -= (y_idx * y_idx)             # <<<<<<<<<<<<<<
  * 
- *             self.mean_left = (self.n_left * self.mean_left + y_idx) / \             # <<<<<<<<<<<<<<
- *                 <double>(self.n_left + 1)
- *             self.mean_right = ((self.n_samples - self.n_left) * \
+ *                 mean_left[k] = (n_left * mean_left[k] + y_idx) / <double>(n_left + 1)
  */
-    __pyx_v_self->mean_left = (((__pyx_v_self->n_left * __pyx_v_self->mean_left) + __pyx_v_y_idx) / ((double)(__pyx_v_self->n_left + 1)));
+      __pyx_t_4 = __pyx_v_k;
+      (__pyx_v_sq_sum_right[__pyx_t_4]) = ((__pyx_v_sq_sum_right[__pyx_t_4]) - (__pyx_v_y_idx * __pyx_v_y_idx));
 
-    /* "sklearn/tree/_tree.pyx":372
- *             self.mean_left = (self.n_left * self.mean_left + y_idx) / \
- *                 <double>(self.n_left + 1)
- *             self.mean_right = ((self.n_samples - self.n_left) * \             # <<<<<<<<<<<<<<
- *                 self.mean_right - y_idx) / \
- *                 <double>(self.n_samples - self.n_left - 1)
+      /* "sklearn/tree/_tree.pyx":546
+ *                 sq_sum_right[k] -= (y_idx * y_idx)
+ * 
+ *                 mean_left[k] = (n_left * mean_left[k] + y_idx) / <double>(n_left + 1)             # <<<<<<<<<<<<<<
+ *                 mean_right[k] = ((n_samples - n_left) * mean_right[k] - y_idx) / <double>(n_samples - n_left - 1)
+ * 
+ */
+      (__pyx_v_mean_left[__pyx_v_k]) = (((__pyx_v_n_left * (__pyx_v_mean_left[__pyx_v_k])) + __pyx_v_y_idx) / ((double)(__pyx_v_n_left + 1)));
+
+      /* "sklearn/tree/_tree.pyx":547
+ * 
+ *                 mean_left[k] = (n_left * mean_left[k] + y_idx) / <double>(n_left + 1)
+ *                 mean_right[k] = ((n_samples - n_left) * mean_right[k] - y_idx) / <double>(n_samples - n_left - 1)             # <<<<<<<<<<<<<<
+ * 
+ *             n_left += 1
+ */
+      (__pyx_v_mean_right[__pyx_v_k]) = ((((__pyx_v_n_samples - __pyx_v_n_left) * (__pyx_v_mean_right[__pyx_v_k])) - __pyx_v_y_idx) / ((double)((__pyx_v_n_samples - __pyx_v_n_left) - 1)));
+    }
+
+    /* "sklearn/tree/_tree.pyx":549
+ *                 mean_right[k] = ((n_samples - n_left) * mean_right[k] - y_idx) / <double>(n_samples - n_left - 1)
+ * 
+ *             n_left += 1             # <<<<<<<<<<<<<<
+ *             self.n_left = n_left
+ *             n_right -= 1
  */
-    __pyx_v_self->mean_right = ((((__pyx_v_self->n_samples - __pyx_v_self->n_left) * __pyx_v_self->mean_right) - __pyx_v_y_idx) / ((double)((__pyx_v_self->n_samples - __pyx_v_self->n_left) - 1)));
+    __pyx_v_n_left = (__pyx_v_n_left + 1);
 
-    /* "sklearn/tree/_tree.pyx":376
- *                 <double>(self.n_samples - self.n_left - 1)
+    /* "sklearn/tree/_tree.pyx":550
  * 
- *             self.n_right -= 1             # <<<<<<<<<<<<<<
- *             self.n_left += 1
+ *             n_left += 1
+ *             self.n_left = n_left             # <<<<<<<<<<<<<<
+ *             n_right -= 1
+ *             self.n_right = n_right
+ */
+    __pyx_v_self->n_left = __pyx_v_n_left;
+
+    /* "sklearn/tree/_tree.pyx":551
+ *             n_left += 1
+ *             self.n_left = n_left
+ *             n_right -= 1             # <<<<<<<<<<<<<<
+ *             self.n_right = n_right
  * 
  */
-    __pyx_v_self->n_right = (__pyx_v_self->n_right - 1);
+    __pyx_v_n_right = (__pyx_v_n_right - 1);
 
-    /* "sklearn/tree/_tree.pyx":377
+    /* "sklearn/tree/_tree.pyx":552
+ *             self.n_left = n_left
+ *             n_right -= 1
+ *             self.n_right = n_right             # <<<<<<<<<<<<<<
  * 
- *             self.n_right -= 1
- *             self.n_left += 1             # <<<<<<<<<<<<<<
+ *             for k from 0 <= k < n_outputs:
+ */
+    __pyx_v_self->n_right = __pyx_v_n_right;
+
+    /* "sklearn/tree/_tree.pyx":554
+ *             self.n_right = n_right
  * 
- *             self.var_left = self.sq_sum_left - \
+ *             for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *                 var_left[k] = sq_sum_left[k] - n_left * (mean_left[k] * mean_left[k])
+ *                 var_right[k] = sq_sum_right[k] - n_right * (mean_right[k] * mean_right[k])
  */
-    __pyx_v_self->n_left = (__pyx_v_self->n_left + 1);
+    __pyx_t_3 = __pyx_v_n_outputs;
+    for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_3; __pyx_v_k++) {
 
-    /* "sklearn/tree/_tree.pyx":379
- *             self.n_left += 1
+      /* "sklearn/tree/_tree.pyx":555
+ * 
+ *             for k from 0 <= k < n_outputs:
+ *                 var_left[k] = sq_sum_left[k] - n_left * (mean_left[k] * mean_left[k])             # <<<<<<<<<<<<<<
+ *                 var_right[k] = sq_sum_right[k] - n_right * (mean_right[k] * mean_right[k])
  * 
- *             self.var_left = self.sq_sum_left - \             # <<<<<<<<<<<<<<
- *                 self.n_left * (self.mean_left * self.mean_left)
- *             self.var_right = self.sq_sum_right - \
  */
-    __pyx_v_self->var_left = (__pyx_v_self->sq_sum_left - (__pyx_v_self->n_left * (__pyx_v_self->mean_left * __pyx_v_self->mean_left)));
+      (__pyx_v_var_left[__pyx_v_k]) = ((__pyx_v_sq_sum_left[__pyx_v_k]) - (__pyx_v_n_left * ((__pyx_v_mean_left[__pyx_v_k]) * (__pyx_v_mean_left[__pyx_v_k]))));
 
-    /* "sklearn/tree/_tree.pyx":381
- *             self.var_left = self.sq_sum_left - \
- *                 self.n_left * (self.mean_left * self.mean_left)
- *             self.var_right = self.sq_sum_right - \             # <<<<<<<<<<<<<<
- *                 self.n_right * (self.mean_right * self.mean_right)
+      /* "sklearn/tree/_tree.pyx":556
+ *             for k from 0 <= k < n_outputs:
+ *                 var_left[k] = sq_sum_left[k] - n_left * (mean_left[k] * mean_left[k])
+ *                 var_right[k] = sq_sum_right[k] - n_right * (mean_right[k] * mean_right[k])             # <<<<<<<<<<<<<<
  * 
+ *         return n_left
  */
-    __pyx_v_self->var_right = (__pyx_v_self->sq_sum_right - (__pyx_v_self->n_right * (__pyx_v_self->mean_right * __pyx_v_self->mean_right)));
+      (__pyx_v_var_right[__pyx_v_k]) = ((__pyx_v_sq_sum_right[__pyx_v_k]) - (__pyx_v_n_right * ((__pyx_v_mean_right[__pyx_v_k]) * (__pyx_v_mean_right[__pyx_v_k]))));
+    }
     __pyx_L3_continue:;
   }
 
-  /* "sklearn/tree/_tree.pyx":384
- *                 self.n_right * (self.mean_right * self.mean_right)
+  /* "sklearn/tree/_tree.pyx":558
+ *                 var_right[k] = sq_sum_right[k] - n_right * (mean_right[k] * mean_right[k])
  * 
- *         return self.n_left             # <<<<<<<<<<<<<<
+ *         return n_left             # <<<<<<<<<<<<<<
  * 
  *     cdef double eval(self):
  */
-  __pyx_r = __pyx_v_self->n_left;
+  __pyx_r = __pyx_v_n_left;
   goto __pyx_L0;
 
   __pyx_r = 0;
@@ -3349,8 +4512,8 @@ static int __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_update(struct __p
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":386
- *         return self.n_left
+/* "sklearn/tree/_tree.pyx":560
+ *         return n_left
  * 
  *     cdef double eval(self):             # <<<<<<<<<<<<<<
  *         pass
@@ -3367,36 +4530,51 @@ static double __pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_eval(CYTHON_UN
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":389
+/* "sklearn/tree/_tree.pyx":563
  *         pass
  * 
  *     cpdef np.ndarray init_value(self):             # <<<<<<<<<<<<<<
- *         ## TODO is calling np.asarray a performance issue?
- *         return np.asarray(self.mean_init)
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef double* mean_init = self.mean_init
  */
 
-static PyObject *__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_3init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_5init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
 static PyArrayObject *__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init_value(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self, int __pyx_skip_dispatch) {
+  int __pyx_v_n_outputs;
+  double *__pyx_v_mean_init;
+  PyArrayObject *__pyx_v_value = 0;
+  int __pyx_v_k;
+  __Pyx_LocalBuf_ND __pyx_pybuffernd_value;
+  __Pyx_Buffer __pyx_pybuffer_value;
   PyArrayObject *__pyx_r = NULL;
   __Pyx_RefNannyDeclarations
   PyObject *__pyx_t_1 = NULL;
   PyObject *__pyx_t_2 = NULL;
   PyObject *__pyx_t_3 = NULL;
+  PyObject *__pyx_t_4 = NULL;
+  PyArrayObject *__pyx_t_5 = NULL;
+  int __pyx_t_6;
+  int __pyx_t_7;
+  long __pyx_t_8;
   int __pyx_lineno = 0;
   const char *__pyx_filename = NULL;
   int __pyx_clineno = 0;
   __Pyx_RefNannySetupContext("init_value", 0);
+  __pyx_pybuffer_value.pybuffer.buf = NULL;
+  __pyx_pybuffer_value.refcount = 0;
+  __pyx_pybuffernd_value.data = NULL;
+  __pyx_pybuffernd_value.rcbuffer = &__pyx_pybuffer_value;
   /* Check if called by wrapper */
   if (unlikely(__pyx_skip_dispatch)) ;
   /* Check if overriden in Python */
   else if (unlikely(Py_TYPE(((PyObject *)__pyx_v_self))->tp_dictoffset != 0)) {
-    __pyx_t_1 = PyObject_GetAttr(((PyObject *)__pyx_v_self), __pyx_n_s__init_value); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 389; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_1 = PyObject_GetAttr(((PyObject *)__pyx_v_self), __pyx_n_s__init_value); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 563; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     __Pyx_GOTREF(__pyx_t_1);
-    if (!PyCFunction_Check(__pyx_t_1) || (PyCFunction_GET_FUNCTION(__pyx_t_1) != (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_3init_value)) {
+    if (!PyCFunction_Check(__pyx_t_1) || (PyCFunction_GET_FUNCTION(__pyx_t_1) != (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_5init_value)) {
       __Pyx_XDECREF(((PyObject *)__pyx_r));
-      __pyx_t_2 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_empty_tuple), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 389; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      __pyx_t_2 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_empty_tuple), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 563; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
       __Pyx_GOTREF(__pyx_t_2);
-      if (!(likely(((__pyx_t_2) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_2, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 389; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      if (!(likely(((__pyx_t_2) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_2, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 563; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
       __pyx_r = ((PyArrayObject *)__pyx_t_2);
       __pyx_t_2 = 0;
       __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
@@ -3405,33 +4583,108 @@ static PyArrayObject *__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init_v
     __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
   }
 
-  /* "sklearn/tree/_tree.pyx":391
+  /* "sklearn/tree/_tree.pyx":564
+ * 
+ *     cpdef np.ndarray init_value(self):
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ *         cdef double* mean_init = self.mean_init
+ * 
+ */
+  __pyx_v_n_outputs = __pyx_v_self->n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":565
  *     cpdef np.ndarray init_value(self):
- *         ## TODO is calling np.asarray a performance issue?
- *         return np.asarray(self.mean_init)             # <<<<<<<<<<<<<<
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef double* mean_init = self.mean_init             # <<<<<<<<<<<<<<
+ * 
+ *         cdef np.ndarray[DTYPE_t, ndim=2] value = np.zeros((n_outputs, 1), dtype=DTYPE)
+ */
+  __pyx_v_mean_init = __pyx_v_self->mean_init;
+
+  /* "sklearn/tree/_tree.pyx":567
+ *         cdef double* mean_init = self.mean_init
  * 
+ *         cdef np.ndarray[DTYPE_t, ndim=2] value = np.zeros((n_outputs, 1), dtype=DTYPE)             # <<<<<<<<<<<<<<
+ *         cdef int k
  * 
  */
-  __Pyx_XDECREF(((PyObject *)__pyx_r));
-  __pyx_t_1 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 391; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_2 = PyObject_GetAttr(__pyx_t_1, __pyx_n_s__asarray); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 391; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_2 = PyObject_GetAttr(__pyx_t_1, __pyx_n_s__zeros); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_2);
   __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __pyx_t_1 = PyFloat_FromDouble(__pyx_v_self->mean_init); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 391; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = PyInt_FromLong(__pyx_v_n_outputs); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_3 = PyTuple_New(1); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 391; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_3 = PyTuple_New(2); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_3);
   PyTuple_SET_ITEM(__pyx_t_3, 0, __pyx_t_1);
   __Pyx_GIVEREF(__pyx_t_1);
+  __Pyx_INCREF(__pyx_int_1);
+  PyTuple_SET_ITEM(__pyx_t_3, 1, __pyx_int_1);
+  __Pyx_GIVEREF(__pyx_int_1);
   __pyx_t_1 = 0;
-  __pyx_t_1 = PyObject_Call(__pyx_t_2, ((PyObject *)__pyx_t_3), NULL); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 391; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = PyTuple_New(1); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
+  PyTuple_SET_ITEM(__pyx_t_1, 0, ((PyObject *)__pyx_t_3));
+  __Pyx_GIVEREF(((PyObject *)__pyx_t_3));
+  __pyx_t_3 = 0;
+  __pyx_t_3 = PyDict_New(); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(((PyObject *)__pyx_t_3));
+  __pyx_t_4 = __Pyx_GetName(__pyx_m, __pyx_n_s__DTYPE); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_4);
+  if (PyDict_SetItem(__pyx_t_3, ((PyObject *)__pyx_n_s__dtype), __pyx_t_4) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_4); __pyx_t_4 = 0;
+  __pyx_t_4 = PyObject_Call(__pyx_t_2, ((PyObject *)__pyx_t_1), ((PyObject *)__pyx_t_3)); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_4);
   __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+  __Pyx_DECREF(((PyObject *)__pyx_t_1)); __pyx_t_1 = 0;
   __Pyx_DECREF(((PyObject *)__pyx_t_3)); __pyx_t_3 = 0;
-  if (!(likely(((__pyx_t_1) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_1, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 391; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __pyx_r = ((PyArrayObject *)__pyx_t_1);
-  __pyx_t_1 = 0;
+  if (!(likely(((__pyx_t_4) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_4, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_5 = ((PyArrayObject *)__pyx_t_4);
+  {
+    __Pyx_BufFmt_StackElem __pyx_stack[1];
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_value.rcbuffer->pybuffer, (PyObject*)__pyx_t_5, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_STRIDES| PyBUF_WRITABLE, 2, 0, __pyx_stack) == -1)) {
+      __pyx_v_value = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None); __pyx_pybuffernd_value.rcbuffer->pybuffer.buf = NULL;
+      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 567; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    } else {__pyx_pybuffernd_value.diminfo[0].strides = __pyx_pybuffernd_value.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_value.diminfo[0].shape = __pyx_pybuffernd_value.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_value.diminfo[1].strides = __pyx_pybuffernd_value.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_value.diminfo[1].shape = __pyx_pybuffernd_value.rcbuffer->pybuffer.shape[1];
+    }
+  }
+  __pyx_t_5 = 0;
+  __pyx_v_value = ((PyArrayObject *)__pyx_t_4);
+  __pyx_t_4 = 0;
+
+  /* "sklearn/tree/_tree.pyx":570
+ *         cdef int k
+ * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             value[k, 0] = <DTYPE_t> (mean_init[k])
+ * 
+ */
+  __pyx_t_6 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_6; __pyx_v_k++) {
+
+    /* "sklearn/tree/_tree.pyx":571
+ * 
+ *         for k from 0 <= k < n_outputs:
+ *             value[k, 0] = <DTYPE_t> (mean_init[k])             # <<<<<<<<<<<<<<
+ * 
+ *         return value
+ */
+    __pyx_t_7 = __pyx_v_k;
+    __pyx_t_8 = 0;
+    *__Pyx_BufPtrStrided2d(__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_pybuffernd_value.rcbuffer->pybuffer.buf, __pyx_t_7, __pyx_pybuffernd_value.diminfo[0].strides, __pyx_t_8, __pyx_pybuffernd_value.diminfo[1].strides) = ((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t)(__pyx_v_mean_init[__pyx_v_k]));
+  }
+
+  /* "sklearn/tree/_tree.pyx":573
+ *             value[k, 0] = <DTYPE_t> (mean_init[k])
+ * 
+ *         return value             # <<<<<<<<<<<<<<
+ * 
+ * 
+ */
+  __Pyx_XDECREF(((PyObject *)__pyx_r));
+  __Pyx_INCREF(((PyObject *)__pyx_v_value));
+  __pyx_r = ((PyArrayObject *)__pyx_v_value);
   goto __pyx_L0;
 
   __pyx_r = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None);
@@ -3440,34 +4693,43 @@ static PyArrayObject *__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init_v
   __Pyx_XDECREF(__pyx_t_1);
   __Pyx_XDECREF(__pyx_t_2);
   __Pyx_XDECREF(__pyx_t_3);
+  __Pyx_XDECREF(__pyx_t_4);
+  { PyObject *__pyx_type, *__pyx_value, *__pyx_tb;
+    __Pyx_ErrFetch(&__pyx_type, &__pyx_value, &__pyx_tb);
+    __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_value.rcbuffer->pybuffer);
+  __Pyx_ErrRestore(__pyx_type, __pyx_value, __pyx_tb);}
   __Pyx_AddTraceback("sklearn.tree._tree.RegressionCriterion.init_value", __pyx_clineno, __pyx_lineno, __pyx_filename);
   __pyx_r = 0;
+  goto __pyx_L2;
   __pyx_L0:;
+  __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_value.rcbuffer->pybuffer);
+  __pyx_L2:;
+  __Pyx_XDECREF((PyObject *)__pyx_v_value);
   __Pyx_XGIVEREF((PyObject *)__pyx_r);
   __Pyx_RefNannyFinishContext();
   return __pyx_r;
 }
 
 /* Python wrapper */
-static PyObject *__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_3init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
-static PyObject *__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_3init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused) {
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_5init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused); /*proto*/
+static PyObject *__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_5init_value(PyObject *__pyx_v_self, CYTHON_UNUSED PyObject *unused) {
   PyObject *__pyx_r = 0;
   __Pyx_RefNannyDeclarations
   __Pyx_RefNannySetupContext("init_value (wrapper)", 0);
-  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_2init_value(((struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *)__pyx_v_self));
+  __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_4init_value(((struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *)__pyx_v_self));
   __Pyx_RefNannyFinishContext();
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":389
+/* "sklearn/tree/_tree.pyx":563
  *         pass
  * 
  *     cpdef np.ndarray init_value(self):             # <<<<<<<<<<<<<<
- *         ## TODO is calling np.asarray a performance issue?
- *         return np.asarray(self.mean_init)
+ *         cdef int n_outputs = self.n_outputs
+ *         cdef double* mean_init = self.mean_init
  */
 
-static PyObject *__pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_2init_value(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self) {
+static PyObject *__pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_4init_value(struct __pyx_obj_7sklearn_4tree_5_tree_RegressionCriterion *__pyx_v_self) {
   PyObject *__pyx_r = NULL;
   __Pyx_RefNannyDeclarations
   PyObject *__pyx_t_1 = NULL;
@@ -3476,7 +4738,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_2init_valu
   int __pyx_clineno = 0;
   __Pyx_RefNannySetupContext("init_value", 0);
   __Pyx_XDECREF(__pyx_r);
-  __pyx_t_1 = ((PyObject *)((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_RegressionCriterion *)__pyx_v_self->__pyx_base.__pyx_vtab)->__pyx_base.init_value(((struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *)__pyx_v_self), 1)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 389; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = ((PyObject *)((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_RegressionCriterion *)__pyx_v_self->__pyx_base.__pyx_vtab)->__pyx_base.init_value(((struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *)__pyx_v_self), 1)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 563; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
   __pyx_r = __pyx_t_1;
   __pyx_t_1 = 0;
@@ -3494,27 +4756,98 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_19RegressionCriterion_2init_valu
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":400
+/* "sklearn/tree/_tree.pyx":582
  *     """
  * 
  *     cdef double eval(self):             # <<<<<<<<<<<<<<
- *         return self.var_left + self.var_right
- * 
+ *         cdef double* var_left = self.var_left
+ *         cdef double* var_right = self.var_right
  */
 
 static double __pyx_f_7sklearn_4tree_5_tree_3MSE_eval(struct __pyx_obj_7sklearn_4tree_5_tree_MSE *__pyx_v_self) {
+  double *__pyx_v_var_left;
+  double *__pyx_v_var_right;
+  int __pyx_v_n_outputs;
+  int __pyx_v_k;
+  double __pyx_v_total;
   double __pyx_r;
   __Pyx_RefNannyDeclarations
+  int __pyx_t_1;
   __Pyx_RefNannySetupContext("eval", 0);
 
-  /* "sklearn/tree/_tree.pyx":401
+  /* "sklearn/tree/_tree.pyx":583
+ * 
+ *     cdef double eval(self):
+ *         cdef double* var_left = self.var_left             # <<<<<<<<<<<<<<
+ *         cdef double* var_right = self.var_right
  * 
+ */
+  __pyx_v_var_left = __pyx_v_self->__pyx_base.var_left;
+
+  /* "sklearn/tree/_tree.pyx":584
  *     cdef double eval(self):
- *         return self.var_left + self.var_right             # <<<<<<<<<<<<<<
+ *         cdef double* var_left = self.var_left
+ *         cdef double* var_right = self.var_right             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int n_outputs = self.n_outputs
+ */
+  __pyx_v_var_right = __pyx_v_self->__pyx_base.var_right;
+
+  /* "sklearn/tree/_tree.pyx":586
+ *         cdef double* var_right = self.var_right
+ * 
+ *         cdef int n_outputs = self.n_outputs             # <<<<<<<<<<<<<<
+ * 
+ *         cdef int k
+ */
+  __pyx_v_n_outputs = __pyx_v_self->__pyx_base.n_outputs;
+
+  /* "sklearn/tree/_tree.pyx":589
+ * 
+ *         cdef int k
+ *         cdef double total = 0.0             # <<<<<<<<<<<<<<
+ * 
+ *         for k from 0 <= k < n_outputs:
+ */
+  __pyx_v_total = 0.0;
+
+  /* "sklearn/tree/_tree.pyx":591
+ *         cdef double total = 0.0
+ * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             total += var_left[k]
+ *             total += var_right[k]
+ */
+  __pyx_t_1 = __pyx_v_n_outputs;
+  for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_1; __pyx_v_k++) {
+
+    /* "sklearn/tree/_tree.pyx":592
+ * 
+ *         for k from 0 <= k < n_outputs:
+ *             total += var_left[k]             # <<<<<<<<<<<<<<
+ *             total += var_right[k]
+ * 
+ */
+    __pyx_v_total = (__pyx_v_total + (__pyx_v_var_left[__pyx_v_k]));
+
+    /* "sklearn/tree/_tree.pyx":593
+ *         for k from 0 <= k < n_outputs:
+ *             total += var_left[k]
+ *             total += var_right[k]             # <<<<<<<<<<<<<<
+ * 
+ *         return total / n_outputs
+ */
+    __pyx_v_total = (__pyx_v_total + (__pyx_v_var_right[__pyx_v_k]));
+  }
+
+  /* "sklearn/tree/_tree.pyx":595
+ *             total += var_right[k]
+ * 
+ *         return total / n_outputs             # <<<<<<<<<<<<<<
  * 
  * 
  */
-  __pyx_r = (__pyx_v_self->__pyx_base.var_left + __pyx_v_self->__pyx_base.var_right);
+  __pyx_r = (__pyx_v_total / __pyx_v_n_outputs);
   goto __pyx_L0;
 
   __pyx_r = 0;
@@ -3558,17 +4891,17 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_1_random_sample_mask(PyObject *_
         values[1] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__n_total_in_bag);
         if (likely(values[1])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_random_sample_mask", 1, 3, 3, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_random_sample_mask", 1, 3, 3, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  2:
         values[2] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__random_state);
         if (likely(values[2])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_random_sample_mask", 1, 3, 3, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_random_sample_mask", 1, 3, 3, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
       }
       if (unlikely(kw_args > 0)) {
-        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_random_sample_mask") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_random_sample_mask") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
       }
     } else if (PyTuple_GET_SIZE(__pyx_args) != 3) {
       goto __pyx_L5_argtuple_error;
@@ -3577,13 +4910,13 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_1_random_sample_mask(PyObject *_
       values[1] = PyTuple_GET_ITEM(__pyx_args, 1);
       values[2] = PyTuple_GET_ITEM(__pyx_args, 2);
     }
-    __pyx_v_n_total_samples = __Pyx_PyInt_AsInt(values[0]); if (unlikely((__pyx_v_n_total_samples == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
-    __pyx_v_n_total_in_bag = __Pyx_PyInt_AsInt(values[1]); if (unlikely((__pyx_v_n_total_in_bag == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_n_total_samples = __Pyx_PyInt_AsInt(values[0]); if (unlikely((__pyx_v_n_total_samples == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_n_total_in_bag = __Pyx_PyInt_AsInt(values[1]); if (unlikely((__pyx_v_n_total_in_bag == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
     __pyx_v_random_state = values[2];
   }
   goto __pyx_L4_argument_unpacking_done;
   __pyx_L5_argtuple_error:;
-  __Pyx_RaiseArgtupleInvalid("_random_sample_mask", 1, 3, 3, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+  __Pyx_RaiseArgtupleInvalid("_random_sample_mask", 1, 3, 3, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
   __pyx_L3_error:;
   __Pyx_AddTraceback("sklearn.tree._tree._random_sample_mask", __pyx_clineno, __pyx_lineno, __pyx_filename);
   __Pyx_RefNannyFinishContext();
@@ -3594,7 +4927,7 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_1_random_sample_mask(PyObject *_
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":409
+/* "sklearn/tree/_tree.pyx":604
  * 
  * 
  * def _random_sample_mask(int n_total_samples, int n_total_in_bag, random_state):             # <<<<<<<<<<<<<<
@@ -3624,7 +4957,6 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree__random_sample_mask(CYTHON_UNUSE
   int __pyx_t_9;
   int __pyx_t_10;
   int __pyx_t_11;
-  int __pyx_t_12;
   int __pyx_lineno = 0;
   const char *__pyx_filename = NULL;
   int __pyx_clineno = 0;
@@ -3638,33 +4970,33 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree__random_sample_mask(CYTHON_UNUSE
   __pyx_pybuffernd_sample_mask.data = NULL;
   __pyx_pybuffernd_sample_mask.rcbuffer = &__pyx_pybuffer_sample_mask;
 
-  /* "sklearn/tree/_tree.pyx":428
+  /* "sklearn/tree/_tree.pyx":623
  *     """
  *     cdef np.ndarray[np.float64_t, ndim=1, mode="c"] rand = \
  *          random_state.rand(n_total_samples)             # <<<<<<<<<<<<<<
  *     cdef np.ndarray[BOOL_t, ndim=1, mode="c"] sample_mask = \
  *          np.zeros((n_total_samples,), dtype=np.int8)
  */
-  __pyx_t_1 = PyObject_GetAttr(__pyx_v_random_state, __pyx_n_s__rand); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 428; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = PyObject_GetAttr(__pyx_v_random_state, __pyx_n_s__rand); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 623; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_2 = PyInt_FromLong(__pyx_v_n_total_samples); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 428; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_2 = PyInt_FromLong(__pyx_v_n_total_samples); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 623; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_2);
-  __pyx_t_3 = PyTuple_New(1); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 428; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_3 = PyTuple_New(1); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 623; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_3);
   PyTuple_SET_ITEM(__pyx_t_3, 0, __pyx_t_2);
   __Pyx_GIVEREF(__pyx_t_2);
   __pyx_t_2 = 0;
-  __pyx_t_2 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_t_3), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 428; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_2 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_t_3), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 623; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_2);
   __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
   __Pyx_DECREF(((PyObject *)__pyx_t_3)); __pyx_t_3 = 0;
-  if (!(likely(((__pyx_t_2) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_2, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 428; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (!(likely(((__pyx_t_2) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_2, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 623; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_t_4 = ((PyArrayObject *)__pyx_t_2);
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
     if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_rand.rcbuffer->pybuffer, (PyObject*)__pyx_t_4, &__Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {
       __pyx_v_rand = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None); __pyx_pybuffernd_rand.rcbuffer->pybuffer.buf = NULL;
-      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 427; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 622; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     } else {__pyx_pybuffernd_rand.diminfo[0].strides = __pyx_pybuffernd_rand.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_rand.diminfo[0].shape = __pyx_pybuffernd_rand.rcbuffer->pybuffer.shape[0];
     }
   }
@@ -3672,51 +5004,51 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree__random_sample_mask(CYTHON_UNUSE
   __pyx_v_rand = ((PyArrayObject *)__pyx_t_2);
   __pyx_t_2 = 0;
 
-  /* "sklearn/tree/_tree.pyx":430
+  /* "sklearn/tree/_tree.pyx":625
  *          random_state.rand(n_total_samples)
  *     cdef np.ndarray[BOOL_t, ndim=1, mode="c"] sample_mask = \
  *          np.zeros((n_total_samples,), dtype=np.int8)             # <<<<<<<<<<<<<<
  * 
  *     cdef int n_bagged = 0
  */
-  __pyx_t_2 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_2 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_2);
-  __pyx_t_3 = PyObject_GetAttr(__pyx_t_2, __pyx_n_s__zeros); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_3 = PyObject_GetAttr(__pyx_t_2, __pyx_n_s__zeros); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_3);
   __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  __pyx_t_2 = PyInt_FromLong(__pyx_v_n_total_samples); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_2 = PyInt_FromLong(__pyx_v_n_total_samples); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_2);
-  __pyx_t_1 = PyTuple_New(1); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = PyTuple_New(1); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
   PyTuple_SET_ITEM(__pyx_t_1, 0, __pyx_t_2);
   __Pyx_GIVEREF(__pyx_t_2);
   __pyx_t_2 = 0;
-  __pyx_t_2 = PyTuple_New(1); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_2 = PyTuple_New(1); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_2);
   PyTuple_SET_ITEM(__pyx_t_2, 0, ((PyObject *)__pyx_t_1));
   __Pyx_GIVEREF(((PyObject *)__pyx_t_1));
   __pyx_t_1 = 0;
-  __pyx_t_1 = PyDict_New(); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = PyDict_New(); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(((PyObject *)__pyx_t_1));
-  __pyx_t_5 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_5 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_5);
-  __pyx_t_6 = PyObject_GetAttr(__pyx_t_5, __pyx_n_s__int8); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_6 = PyObject_GetAttr(__pyx_t_5, __pyx_n_s__int8); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_6);
   __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
-  if (PyDict_SetItem(__pyx_t_1, ((PyObject *)__pyx_n_s__dtype), __pyx_t_6) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (PyDict_SetItem(__pyx_t_1, ((PyObject *)__pyx_n_s__dtype), __pyx_t_6) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_DECREF(__pyx_t_6); __pyx_t_6 = 0;
-  __pyx_t_6 = PyObject_Call(__pyx_t_3, ((PyObject *)__pyx_t_2), ((PyObject *)__pyx_t_1)); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_6 = PyObject_Call(__pyx_t_3, ((PyObject *)__pyx_t_2), ((PyObject *)__pyx_t_1)); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_6);
   __Pyx_DECREF(__pyx_t_3); __pyx_t_3 = 0;
   __Pyx_DECREF(((PyObject *)__pyx_t_2)); __pyx_t_2 = 0;
   __Pyx_DECREF(((PyObject *)__pyx_t_1)); __pyx_t_1 = 0;
-  if (!(likely(((__pyx_t_6) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_6, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 430; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (!(likely(((__pyx_t_6) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_6, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_t_7 = ((PyArrayObject *)__pyx_t_6);
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
     if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_sample_mask.rcbuffer->pybuffer, (PyObject*)__pyx_t_7, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_BOOL_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS| PyBUF_WRITABLE, 1, 0, __pyx_stack) == -1)) {
       __pyx_v_sample_mask = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None); __pyx_pybuffernd_sample_mask.rcbuffer->pybuffer.buf = NULL;
-      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 429; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 624; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     } else {__pyx_pybuffernd_sample_mask.diminfo[0].strides = __pyx_pybuffernd_sample_mask.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_sample_mask.diminfo[0].shape = __pyx_pybuffernd_sample_mask.rcbuffer->pybuffer.shape[0];
     }
   }
@@ -3724,57 +5056,56 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree__random_sample_mask(CYTHON_UNUSE
   __pyx_v_sample_mask = ((PyArrayObject *)__pyx_t_6);
   __pyx_t_6 = 0;
 
-  /* "sklearn/tree/_tree.pyx":432
+  /* "sklearn/tree/_tree.pyx":627
  *          np.zeros((n_total_samples,), dtype=np.int8)
  * 
  *     cdef int n_bagged = 0             # <<<<<<<<<<<<<<
  *     cdef int i = 0
- *     for i in range(n_total_samples):
+ *     for i from 0 <= i < n_total_samples:
  */
   __pyx_v_n_bagged = 0;
 
-  /* "sklearn/tree/_tree.pyx":433
+  /* "sklearn/tree/_tree.pyx":628
  * 
  *     cdef int n_bagged = 0
  *     cdef int i = 0             # <<<<<<<<<<<<<<
- *     for i in range(n_total_samples):
+ *     for i from 0 <= i < n_total_samples:
  *         if rand[i] * (n_total_samples - i) < (n_total_in_bag - n_bagged):
  */
   __pyx_v_i = 0;
 
-  /* "sklearn/tree/_tree.pyx":434
+  /* "sklearn/tree/_tree.pyx":629
  *     cdef int n_bagged = 0
  *     cdef int i = 0
- *     for i in range(n_total_samples):             # <<<<<<<<<<<<<<
+ *     for i from 0 <= i < n_total_samples:             # <<<<<<<<<<<<<<
  *         if rand[i] * (n_total_samples - i) < (n_total_in_bag - n_bagged):
  *             sample_mask[i] = 1
  */
   __pyx_t_8 = __pyx_v_n_total_samples;
-  for (__pyx_t_9 = 0; __pyx_t_9 < __pyx_t_8; __pyx_t_9+=1) {
-    __pyx_v_i = __pyx_t_9;
+  for (__pyx_v_i = 0; __pyx_v_i < __pyx_t_8; __pyx_v_i++) {
 
-    /* "sklearn/tree/_tree.pyx":435
+    /* "sklearn/tree/_tree.pyx":630
  *     cdef int i = 0
- *     for i in range(n_total_samples):
+ *     for i from 0 <= i < n_total_samples:
  *         if rand[i] * (n_total_samples - i) < (n_total_in_bag - n_bagged):             # <<<<<<<<<<<<<<
  *             sample_mask[i] = 1
  *             n_bagged += 1
  */
-    __pyx_t_10 = __pyx_v_i;
-    __pyx_t_11 = (((*__Pyx_BufPtrCContig1d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_rand.rcbuffer->pybuffer.buf, __pyx_t_10, __pyx_pybuffernd_rand.diminfo[0].strides)) * (__pyx_v_n_total_samples - __pyx_v_i)) < (__pyx_v_n_total_in_bag - __pyx_v_n_bagged));
-    if (__pyx_t_11) {
+    __pyx_t_9 = __pyx_v_i;
+    __pyx_t_10 = (((*__Pyx_BufPtrCContig1d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_rand.rcbuffer->pybuffer.buf, __pyx_t_9, __pyx_pybuffernd_rand.diminfo[0].strides)) * (__pyx_v_n_total_samples - __pyx_v_i)) < (__pyx_v_n_total_in_bag - __pyx_v_n_bagged));
+    if (__pyx_t_10) {
 
-      /* "sklearn/tree/_tree.pyx":436
- *     for i in range(n_total_samples):
+      /* "sklearn/tree/_tree.pyx":631
+ *     for i from 0 <= i < n_total_samples:
  *         if rand[i] * (n_total_samples - i) < (n_total_in_bag - n_bagged):
  *             sample_mask[i] = 1             # <<<<<<<<<<<<<<
  *             n_bagged += 1
  * 
  */
-      __pyx_t_12 = __pyx_v_i;
-      *__Pyx_BufPtrCContig1d(__pyx_t_7sklearn_4tree_5_tree_BOOL_t *, __pyx_pybuffernd_sample_mask.rcbuffer->pybuffer.buf, __pyx_t_12, __pyx_pybuffernd_sample_mask.diminfo[0].strides) = 1;
+      __pyx_t_11 = __pyx_v_i;
+      *__Pyx_BufPtrCContig1d(__pyx_t_7sklearn_4tree_5_tree_BOOL_t *, __pyx_pybuffernd_sample_mask.rcbuffer->pybuffer.buf, __pyx_t_11, __pyx_pybuffernd_sample_mask.diminfo[0].strides) = 1;
 
-      /* "sklearn/tree/_tree.pyx":437
+      /* "sklearn/tree/_tree.pyx":632
  *         if rand[i] * (n_total_samples - i) < (n_total_in_bag - n_bagged):
  *             sample_mask[i] = 1
  *             n_bagged += 1             # <<<<<<<<<<<<<<
@@ -3787,7 +5118,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree__random_sample_mask(CYTHON_UNUSE
     __pyx_L5:;
   }
 
-  /* "sklearn/tree/_tree.pyx":439
+  /* "sklearn/tree/_tree.pyx":634
  *             n_bagged += 1
  * 
  *     return sample_mask.astype(np.bool)             # <<<<<<<<<<<<<<
@@ -3795,19 +5126,19 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree__random_sample_mask(CYTHON_UNUSE
  * 
  */
   __Pyx_XDECREF(__pyx_r);
-  __pyx_t_6 = PyObject_GetAttr(((PyObject *)__pyx_v_sample_mask), __pyx_n_s__astype); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 439; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_6 = PyObject_GetAttr(((PyObject *)__pyx_v_sample_mask), __pyx_n_s__astype); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 634; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_6);
-  __pyx_t_1 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 439; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 634; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_2 = PyObject_GetAttr(__pyx_t_1, __pyx_n_s__bool); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 439; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_2 = PyObject_GetAttr(__pyx_t_1, __pyx_n_s__bool); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 634; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_2);
   __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __pyx_t_1 = PyTuple_New(1); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 439; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = PyTuple_New(1); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 634; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
   PyTuple_SET_ITEM(__pyx_t_1, 0, __pyx_t_2);
   __Pyx_GIVEREF(__pyx_t_2);
   __pyx_t_2 = 0;
-  __pyx_t_2 = PyObject_Call(__pyx_t_6, ((PyObject *)__pyx_t_1), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 439; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_2 = PyObject_Call(__pyx_t_6, ((PyObject *)__pyx_t_1), NULL); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 634; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_2);
   __Pyx_DECREF(__pyx_t_6); __pyx_t_6 = 0;
   __Pyx_DECREF(((PyObject *)__pyx_t_1)); __pyx_t_1 = 0;
@@ -3881,29 +5212,29 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_3_apply_tree(PyObject *__pyx_sel
         values[1] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__children);
         if (likely(values[1])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  2:
         values[2] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__feature);
         if (likely(values[2])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  3:
         values[3] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__threshold);
         if (likely(values[3])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  4:
         values[4] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__out);
         if (likely(values[4])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, 4); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, 4); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
       }
       if (unlikely(kw_args > 0)) {
-        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_apply_tree") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_apply_tree") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
       }
     } else if (PyTuple_GET_SIZE(__pyx_args) != 5) {
       goto __pyx_L5_argtuple_error;
@@ -3922,17 +5253,17 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_3_apply_tree(PyObject *__pyx_sel
   }
   goto __pyx_L4_argument_unpacking_done;
   __pyx_L5_argtuple_error:;
-  __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+  __Pyx_RaiseArgtupleInvalid("_apply_tree", 1, 5, 5, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
   __pyx_L3_error:;
   __Pyx_AddTraceback("sklearn.tree._tree._apply_tree", __pyx_clineno, __pyx_lineno, __pyx_filename);
   __Pyx_RefNannyFinishContext();
   return NULL;
   __pyx_L4_argument_unpacking_done:;
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X), __pyx_ptype_5numpy_ndarray, 1, "X", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_children), __pyx_ptype_5numpy_ndarray, 1, "children", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 443; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_feature), __pyx_ptype_5numpy_ndarray, 1, "feature", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 444; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_threshold), __pyx_ptype_5numpy_ndarray, 1, "threshold", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 445; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_out), __pyx_ptype_5numpy_ndarray, 1, "out", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 446; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X), __pyx_ptype_5numpy_ndarray, 1, "X", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_children), __pyx_ptype_5numpy_ndarray, 1, "children", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 638; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_feature), __pyx_ptype_5numpy_ndarray, 1, "feature", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 639; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_threshold), __pyx_ptype_5numpy_ndarray, 1, "threshold", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 640; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_out), __pyx_ptype_5numpy_ndarray, 1, "out", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 641; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_2_apply_tree(__pyx_self, __pyx_v_X, __pyx_v_children, __pyx_v_feature, __pyx_v_threshold, __pyx_v_out);
   goto __pyx_L0;
   __pyx_L1_error:;
@@ -3942,7 +5273,7 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_3_apply_tree(PyObject *__pyx_sel
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":442
+/* "sklearn/tree/_tree.pyx":637
  * 
  * 
  * def _apply_tree(np.ndarray[DTYPE_t, ndim=2] X,             # <<<<<<<<<<<<<<
@@ -3968,22 +5299,21 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_2_apply_tree(CYTHON_UNUSED PyObj
   __Pyx_RefNannyDeclarations
   int __pyx_t_1;
   int __pyx_t_2;
-  int __pyx_t_3;
-  long __pyx_t_4;
+  long __pyx_t_3;
+  int __pyx_t_4;
   int __pyx_t_5;
-  int __pyx_t_6;
-  long __pyx_t_7;
+  long __pyx_t_6;
+  int __pyx_t_7;
   int __pyx_t_8;
   int __pyx_t_9;
   int __pyx_t_10;
-  int __pyx_t_11;
-  __pyx_t_5numpy_int32_t __pyx_t_12;
+  __pyx_t_5numpy_int32_t __pyx_t_11;
+  int __pyx_t_12;
   int __pyx_t_13;
-  int __pyx_t_14;
-  long __pyx_t_15;
-  int __pyx_t_16;
-  long __pyx_t_17;
-  int __pyx_t_18;
+  long __pyx_t_14;
+  int __pyx_t_15;
+  long __pyx_t_16;
+  int __pyx_t_17;
   int __pyx_lineno = 0;
   const char *__pyx_filename = NULL;
   int __pyx_clineno = 0;
@@ -4010,31 +5340,31 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_2_apply_tree(CYTHON_UNUSED PyObj
   __pyx_pybuffernd_out.rcbuffer = &__pyx_pybuffer_out;
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X.rcbuffer->pybuffer, (PyObject*)__pyx_v_X, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_STRIDES, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X.rcbuffer->pybuffer, (PyObject*)__pyx_v_X, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_STRIDES, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_X.diminfo[0].strides = __pyx_pybuffernd_X.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_X.diminfo[0].shape = __pyx_pybuffernd_X.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_X.diminfo[1].strides = __pyx_pybuffernd_X.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_X.diminfo[1].shape = __pyx_pybuffernd_X.rcbuffer->pybuffer.shape[1];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_children.rcbuffer->pybuffer, (PyObject*)__pyx_v_children, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_children.rcbuffer->pybuffer, (PyObject*)__pyx_v_children, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_children.diminfo[0].strides = __pyx_pybuffernd_children.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_children.diminfo[0].shape = __pyx_pybuffernd_children.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_children.diminfo[1].strides = __pyx_pybuffernd_children.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_children.diminfo[1].shape = __pyx_pybuffernd_children.rcbuffer->pybuffer.shape[1];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_feature.rcbuffer->pybuffer, (PyObject*)__pyx_v_feature, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_feature.rcbuffer->pybuffer, (PyObject*)__pyx_v_feature, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_feature.diminfo[0].strides = __pyx_pybuffernd_feature.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_feature.diminfo[0].shape = __pyx_pybuffernd_feature.rcbuffer->pybuffer.shape[0];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_threshold.rcbuffer->pybuffer, (PyObject*)__pyx_v_threshold, &__Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_threshold.rcbuffer->pybuffer, (PyObject*)__pyx_v_threshold, &__Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_threshold.diminfo[0].strides = __pyx_pybuffernd_threshold.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_threshold.diminfo[0].shape = __pyx_pybuffernd_threshold.rcbuffer->pybuffer.shape[0];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_out.rcbuffer->pybuffer, (PyObject*)__pyx_v_out, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES| PyBUF_WRITABLE, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_out.rcbuffer->pybuffer, (PyObject*)__pyx_v_out, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES| PyBUF_WRITABLE, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_out.diminfo[0].strides = __pyx_pybuffernd_out.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_out.diminfo[0].shape = __pyx_pybuffernd_out.rcbuffer->pybuffer.shape[0];
 
-  /* "sklearn/tree/_tree.pyx":449
+  /* "sklearn/tree/_tree.pyx":644
  *     """Finds the terminal region (=leaf node) for each sample in
  *     `X` and sets the corresponding element in `out` to its node id."""
  *     cdef int i = 0             # <<<<<<<<<<<<<<
@@ -4043,45 +5373,44 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_2_apply_tree(CYTHON_UNUSED PyObj
  */
   __pyx_v_i = 0;
 
-  /* "sklearn/tree/_tree.pyx":450
+  /* "sklearn/tree/_tree.pyx":645
  *     `X` and sets the corresponding element in `out` to its node id."""
  *     cdef int i = 0
  *     cdef int n = X.shape[0]             # <<<<<<<<<<<<<<
  *     cdef int node_id = 0
- *     for i in xrange(n):
+ *     for i from 0 <= i < n:
  */
   __pyx_v_n = (__pyx_v_X->dimensions[0]);
 
-  /* "sklearn/tree/_tree.pyx":451
+  /* "sklearn/tree/_tree.pyx":646
  *     cdef int i = 0
  *     cdef int n = X.shape[0]
  *     cdef int node_id = 0             # <<<<<<<<<<<<<<
- *     for i in xrange(n):
+ *     for i from 0 <= i < n:
  *         node_id = 0
  */
   __pyx_v_node_id = 0;
 
-  /* "sklearn/tree/_tree.pyx":452
+  /* "sklearn/tree/_tree.pyx":647
  *     cdef int n = X.shape[0]
  *     cdef int node_id = 0
- *     for i in xrange(n):             # <<<<<<<<<<<<<<
+ *     for i from 0 <= i < n:             # <<<<<<<<<<<<<<
  *         node_id = 0
  *         # While node_id not a leaf
  */
   __pyx_t_1 = __pyx_v_n;
-  for (__pyx_t_2 = 0; __pyx_t_2 < __pyx_t_1; __pyx_t_2+=1) {
-    __pyx_v_i = __pyx_t_2;
+  for (__pyx_v_i = 0; __pyx_v_i < __pyx_t_1; __pyx_v_i++) {
 
-    /* "sklearn/tree/_tree.pyx":453
+    /* "sklearn/tree/_tree.pyx":648
  *     cdef int node_id = 0
- *     for i in xrange(n):
+ *     for i from 0 <= i < n:
  *         node_id = 0             # <<<<<<<<<<<<<<
  *         # While node_id not a leaf
  *         while children[node_id, 0] != -1 and children[node_id, 1] != -1:
  */
     __pyx_v_node_id = 0;
 
-    /* "sklearn/tree/_tree.pyx":455
+    /* "sklearn/tree/_tree.pyx":650
  *         node_id = 0
  *         # While node_id not a leaf
  *         while children[node_id, 0] != -1 and children[node_id, 1] != -1:             # <<<<<<<<<<<<<<
@@ -4089,70 +5418,70 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_2_apply_tree(CYTHON_UNUSED PyObj
  *                 node_id = children[node_id, 0]
  */
     while (1) {
-      __pyx_t_3 = __pyx_v_node_id;
-      __pyx_t_4 = 0;
-      __pyx_t_5 = ((*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_3, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_4, __pyx_pybuffernd_children.diminfo[1].strides)) != -1);
-      if (__pyx_t_5) {
-        __pyx_t_6 = __pyx_v_node_id;
-        __pyx_t_7 = 1;
-        __pyx_t_8 = ((*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_6, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_7, __pyx_pybuffernd_children.diminfo[1].strides)) != -1);
-        __pyx_t_9 = __pyx_t_8;
+      __pyx_t_2 = __pyx_v_node_id;
+      __pyx_t_3 = 0;
+      __pyx_t_4 = ((*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_2, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_3, __pyx_pybuffernd_children.diminfo[1].strides)) != -1);
+      if (__pyx_t_4) {
+        __pyx_t_5 = __pyx_v_node_id;
+        __pyx_t_6 = 1;
+        __pyx_t_7 = ((*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_5, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_6, __pyx_pybuffernd_children.diminfo[1].strides)) != -1);
+        __pyx_t_8 = __pyx_t_7;
       } else {
-        __pyx_t_9 = __pyx_t_5;
+        __pyx_t_8 = __pyx_t_4;
       }
-      if (!__pyx_t_9) break;
+      if (!__pyx_t_8) break;
 
-      /* "sklearn/tree/_tree.pyx":456
+      /* "sklearn/tree/_tree.pyx":651
  *         # While node_id not a leaf
  *         while children[node_id, 0] != -1 and children[node_id, 1] != -1:
  *             if X[i, feature[node_id]] <= threshold[node_id]:             # <<<<<<<<<<<<<<
  *                 node_id = children[node_id, 0]
  *             else:
  */
-      __pyx_t_10 = __pyx_v_node_id;
-      __pyx_t_11 = __pyx_v_i;
-      __pyx_t_12 = (*__Pyx_BufPtrStrided1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_feature.rcbuffer->pybuffer.buf, __pyx_t_10, __pyx_pybuffernd_feature.diminfo[0].strides));
-      __pyx_t_13 = __pyx_v_node_id;
-      __pyx_t_9 = ((*__Pyx_BufPtrStrided2d(__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_pybuffernd_X.rcbuffer->pybuffer.buf, __pyx_t_11, __pyx_pybuffernd_X.diminfo[0].strides, __pyx_t_12, __pyx_pybuffernd_X.diminfo[1].strides)) <= (*__Pyx_BufPtrStrided1d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_threshold.rcbuffer->pybuffer.buf, __pyx_t_13, __pyx_pybuffernd_threshold.diminfo[0].strides)));
-      if (__pyx_t_9) {
+      __pyx_t_9 = __pyx_v_node_id;
+      __pyx_t_10 = __pyx_v_i;
+      __pyx_t_11 = (*__Pyx_BufPtrStrided1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_feature.rcbuffer->pybuffer.buf, __pyx_t_9, __pyx_pybuffernd_feature.diminfo[0].strides));
+      __pyx_t_12 = __pyx_v_node_id;
+      __pyx_t_8 = ((*__Pyx_BufPtrStrided2d(__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_pybuffernd_X.rcbuffer->pybuffer.buf, __pyx_t_10, __pyx_pybuffernd_X.diminfo[0].strides, __pyx_t_11, __pyx_pybuffernd_X.diminfo[1].strides)) <= (*__Pyx_BufPtrStrided1d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_threshold.rcbuffer->pybuffer.buf, __pyx_t_12, __pyx_pybuffernd_threshold.diminfo[0].strides)));
+      if (__pyx_t_8) {
 
-        /* "sklearn/tree/_tree.pyx":457
+        /* "sklearn/tree/_tree.pyx":652
  *         while children[node_id, 0] != -1 and children[node_id, 1] != -1:
  *             if X[i, feature[node_id]] <= threshold[node_id]:
  *                 node_id = children[node_id, 0]             # <<<<<<<<<<<<<<
  *             else:
  *                 node_id = children[node_id, 1]
  */
-        __pyx_t_14 = __pyx_v_node_id;
-        __pyx_t_15 = 0;
-        __pyx_v_node_id = (*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_14, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_15, __pyx_pybuffernd_children.diminfo[1].strides));
+        __pyx_t_13 = __pyx_v_node_id;
+        __pyx_t_14 = 0;
+        __pyx_v_node_id = (*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_13, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_14, __pyx_pybuffernd_children.diminfo[1].strides));
         goto __pyx_L7;
       }
       /*else*/ {
 
-        /* "sklearn/tree/_tree.pyx":459
+        /* "sklearn/tree/_tree.pyx":654
  *                 node_id = children[node_id, 0]
  *             else:
  *                 node_id = children[node_id, 1]             # <<<<<<<<<<<<<<
  *         out[i] = node_id
  * 
  */
-        __pyx_t_16 = __pyx_v_node_id;
-        __pyx_t_17 = 1;
-        __pyx_v_node_id = (*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_16, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_17, __pyx_pybuffernd_children.diminfo[1].strides));
+        __pyx_t_15 = __pyx_v_node_id;
+        __pyx_t_16 = 1;
+        __pyx_v_node_id = (*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_15, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_16, __pyx_pybuffernd_children.diminfo[1].strides));
       }
       __pyx_L7:;
     }
 
-    /* "sklearn/tree/_tree.pyx":460
+    /* "sklearn/tree/_tree.pyx":655
  *             else:
  *                 node_id = children[node_id, 1]
  *         out[i] = node_id             # <<<<<<<<<<<<<<
  * 
  * 
  */
-    __pyx_t_18 = __pyx_v_i;
-    *__Pyx_BufPtrStrided1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_out.rcbuffer->pybuffer.buf, __pyx_t_18, __pyx_pybuffernd_out.diminfo[0].strides) = __pyx_v_node_id;
+    __pyx_t_17 = __pyx_v_i;
+    *__Pyx_BufPtrStrided1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_out.rcbuffer->pybuffer.buf, __pyx_t_17, __pyx_pybuffernd_out.diminfo[0].strides) = __pyx_v_node_id;
   }
 
   __pyx_r = Py_None; __Pyx_INCREF(Py_None);
@@ -4222,35 +5551,35 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_5_predict_tree(PyObject *__pyx_s
         values[1] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__children);
         if (likely(values[1])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  2:
         values[2] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__feature);
         if (likely(values[2])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  3:
         values[3] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__threshold);
         if (likely(values[3])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  4:
         values[4] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__values);
         if (likely(values[4])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 4); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 4); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  5:
         values[5] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__pred);
         if (likely(values[5])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 5); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, 5); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
       }
       if (unlikely(kw_args > 0)) {
-        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_predict_tree") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_predict_tree") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
       }
     } else if (PyTuple_GET_SIZE(__pyx_args) != 6) {
       goto __pyx_L5_argtuple_error;
@@ -4271,18 +5600,18 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_5_predict_tree(PyObject *__pyx_s
   }
   goto __pyx_L4_argument_unpacking_done;
   __pyx_L5_argtuple_error:;
-  __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+  __Pyx_RaiseArgtupleInvalid("_predict_tree", 1, 6, 6, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
   __pyx_L3_error:;
   __Pyx_AddTraceback("sklearn.tree._tree._predict_tree", __pyx_clineno, __pyx_lineno, __pyx_filename);
   __Pyx_RefNannyFinishContext();
   return NULL;
   __pyx_L4_argument_unpacking_done:;
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X), __pyx_ptype_5numpy_ndarray, 1, "X", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_children), __pyx_ptype_5numpy_ndarray, 1, "children", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 464; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_feature), __pyx_ptype_5numpy_ndarray, 1, "feature", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 465; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_threshold), __pyx_ptype_5numpy_ndarray, 1, "threshold", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 466; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_values), __pyx_ptype_5numpy_ndarray, 1, "values", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 467; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_pred), __pyx_ptype_5numpy_ndarray, 1, "pred", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 468; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X), __pyx_ptype_5numpy_ndarray, 1, "X", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_children), __pyx_ptype_5numpy_ndarray, 1, "children", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 659; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_feature), __pyx_ptype_5numpy_ndarray, 1, "feature", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 660; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_threshold), __pyx_ptype_5numpy_ndarray, 1, "threshold", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 661; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_values), __pyx_ptype_5numpy_ndarray, 1, "values", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 662; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_pred), __pyx_ptype_5numpy_ndarray, 1, "pred", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 663; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_4_predict_tree(__pyx_self, __pyx_v_X, __pyx_v_children, __pyx_v_feature, __pyx_v_threshold, __pyx_v_values, __pyx_v_pred);
   goto __pyx_L0;
   __pyx_L1_error:;
@@ -4292,7 +5621,7 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_5_predict_tree(PyObject *__pyx_s
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":463
+/* "sklearn/tree/_tree.pyx":658
  * 
  * 
  * def _predict_tree(np.ndarray[DTYPE_t, ndim=2] X,             # <<<<<<<<<<<<<<
@@ -4302,10 +5631,12 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_5_predict_tree(PyObject *__pyx_s
 
 static PyObject *__pyx_pf_7sklearn_4tree_5_tree_4_predict_tree(CYTHON_UNUSED PyObject *__pyx_self, PyArrayObject *__pyx_v_X, PyArrayObject *__pyx_v_children, PyArrayObject *__pyx_v_feature, PyArrayObject *__pyx_v_threshold, PyArrayObject *__pyx_v_values, PyArrayObject *__pyx_v_pred) {
   int __pyx_v_i;
+  int __pyx_v_k;
+  int __pyx_v_c;
   int __pyx_v_n;
   int __pyx_v_node_id;
-  int __pyx_v_K;
-  int __pyx_v_k;
+  int __pyx_v_n_outputs;
+  int __pyx_v_n_classes;
   __Pyx_LocalBuf_ND __pyx_pybuffernd_X;
   __Pyx_Buffer __pyx_pybuffer_X;
   __Pyx_LocalBuf_ND __pyx_pybuffernd_children;
@@ -4322,27 +5653,28 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_4_predict_tree(CYTHON_UNUSED PyO
   __Pyx_RefNannyDeclarations
   int __pyx_t_1;
   int __pyx_t_2;
-  int __pyx_t_3;
-  long __pyx_t_4;
+  long __pyx_t_3;
+  int __pyx_t_4;
   int __pyx_t_5;
-  int __pyx_t_6;
-  long __pyx_t_7;
+  long __pyx_t_6;
+  int __pyx_t_7;
   int __pyx_t_8;
   int __pyx_t_9;
   int __pyx_t_10;
-  int __pyx_t_11;
-  __pyx_t_5numpy_int32_t __pyx_t_12;
+  __pyx_t_5numpy_int32_t __pyx_t_11;
+  int __pyx_t_12;
   int __pyx_t_13;
-  int __pyx_t_14;
-  long __pyx_t_15;
-  int __pyx_t_16;
-  long __pyx_t_17;
+  long __pyx_t_14;
+  int __pyx_t_15;
+  long __pyx_t_16;
+  int __pyx_t_17;
   int __pyx_t_18;
   int __pyx_t_19;
   int __pyx_t_20;
   int __pyx_t_21;
   int __pyx_t_22;
   int __pyx_t_23;
+  int __pyx_t_24;
   int __pyx_lineno = 0;
   const char *__pyx_filename = NULL;
   int __pyx_clineno = 0;
@@ -4373,92 +5705,91 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_4_predict_tree(CYTHON_UNUSED PyO
   __pyx_pybuffernd_pred.rcbuffer = &__pyx_pybuffer_pred;
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X.rcbuffer->pybuffer, (PyObject*)__pyx_v_X, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_STRIDES, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X.rcbuffer->pybuffer, (PyObject*)__pyx_v_X, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_STRIDES, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_X.diminfo[0].strides = __pyx_pybuffernd_X.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_X.diminfo[0].shape = __pyx_pybuffernd_X.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_X.diminfo[1].strides = __pyx_pybuffernd_X.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_X.diminfo[1].shape = __pyx_pybuffernd_X.rcbuffer->pybuffer.shape[1];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_children.rcbuffer->pybuffer, (PyObject*)__pyx_v_children, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_children.rcbuffer->pybuffer, (PyObject*)__pyx_v_children, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_children.diminfo[0].strides = __pyx_pybuffernd_children.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_children.diminfo[0].shape = __pyx_pybuffernd_children.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_children.diminfo[1].strides = __pyx_pybuffernd_children.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_children.diminfo[1].shape = __pyx_pybuffernd_children.rcbuffer->pybuffer.shape[1];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_feature.rcbuffer->pybuffer, (PyObject*)__pyx_v_feature, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_feature.rcbuffer->pybuffer, (PyObject*)__pyx_v_feature, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_feature.diminfo[0].strides = __pyx_pybuffernd_feature.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_feature.diminfo[0].shape = __pyx_pybuffernd_feature.rcbuffer->pybuffer.shape[0];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_threshold.rcbuffer->pybuffer, (PyObject*)__pyx_v_threshold, &__Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_threshold.rcbuffer->pybuffer, (PyObject*)__pyx_v_threshold, &__Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t, PyBUF_FORMAT| PyBUF_STRIDES, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_threshold.diminfo[0].strides = __pyx_pybuffernd_threshold.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_threshold.diminfo[0].shape = __pyx_pybuffernd_threshold.rcbuffer->pybuffer.shape[0];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_values.rcbuffer->pybuffer, (PyObject*)__pyx_v_values, &__Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t, PyBUF_FORMAT| PyBUF_STRIDES, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_values.rcbuffer->pybuffer, (PyObject*)__pyx_v_values, &__Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t, PyBUF_FORMAT| PyBUF_STRIDES, 3, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
-  __pyx_pybuffernd_values.diminfo[0].strides = __pyx_pybuffernd_values.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_values.diminfo[0].shape = __pyx_pybuffernd_values.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_values.diminfo[1].strides = __pyx_pybuffernd_values.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_values.diminfo[1].shape = __pyx_pybuffernd_values.rcbuffer->pybuffer.shape[1];
+  __pyx_pybuffernd_values.diminfo[0].strides = __pyx_pybuffernd_values.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_values.diminfo[0].shape = __pyx_pybuffernd_values.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_values.diminfo[1].strides = __pyx_pybuffernd_values.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_values.diminfo[1].shape = __pyx_pybuffernd_values.rcbuffer->pybuffer.shape[1]; __pyx_pybuffernd_values.diminfo[2].strides = __pyx_pybuffernd_values.rcbuffer->pybuffer.strides[2]; __pyx_pybuffernd_values.diminfo[2].shape = __pyx_pybuffernd_values.rcbuffer->pybuffer.shape[2];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_pred.rcbuffer->pybuffer, (PyObject*)__pyx_v_pred, &__Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t, PyBUF_FORMAT| PyBUF_STRIDES| PyBUF_WRITABLE, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_pred.rcbuffer->pybuffer, (PyObject*)__pyx_v_pred, &__Pyx_TypeInfo_nn___pyx_t_5numpy_float64_t, PyBUF_FORMAT| PyBUF_STRIDES| PyBUF_WRITABLE, 3, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
-  __pyx_pybuffernd_pred.diminfo[0].strides = __pyx_pybuffernd_pred.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_pred.diminfo[0].shape = __pyx_pybuffernd_pred.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_pred.diminfo[1].strides = __pyx_pybuffernd_pred.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_pred.diminfo[1].shape = __pyx_pybuffernd_pred.rcbuffer->pybuffer.shape[1];
-
-  /* "sklearn/tree/_tree.pyx":470
- *                   np.ndarray[np.float64_t, ndim=2] pred):
- *     """Finds the terminal region (=leaf node) values for each sample. """
- *     cdef int i = 0             # <<<<<<<<<<<<<<
- *     cdef int n = X.shape[0]
- *     cdef int node_id = 0
- */
-  __pyx_v_i = 0;
+  __pyx_pybuffernd_pred.diminfo[0].strides = __pyx_pybuffernd_pred.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_pred.diminfo[0].shape = __pyx_pybuffernd_pred.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_pred.diminfo[1].strides = __pyx_pybuffernd_pred.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_pred.diminfo[1].shape = __pyx_pybuffernd_pred.rcbuffer->pybuffer.shape[1]; __pyx_pybuffernd_pred.diminfo[2].strides = __pyx_pybuffernd_pred.rcbuffer->pybuffer.strides[2]; __pyx_pybuffernd_pred.diminfo[2].shape = __pyx_pybuffernd_pred.rcbuffer->pybuffer.shape[2];
 
-  /* "sklearn/tree/_tree.pyx":471
+  /* "sklearn/tree/_tree.pyx":666
  *     """Finds the terminal region (=leaf node) values for each sample. """
- *     cdef int i = 0
+ *     cdef int i, k, c
  *     cdef int n = X.shape[0]             # <<<<<<<<<<<<<<
  *     cdef int node_id = 0
- *     cdef int K = values.shape[1]
+ *     cdef int n_outputs = values.shape[1]
  */
   __pyx_v_n = (__pyx_v_X->dimensions[0]);
 
-  /* "sklearn/tree/_tree.pyx":472
- *     cdef int i = 0
+  /* "sklearn/tree/_tree.pyx":667
+ *     cdef int i, k, c
  *     cdef int n = X.shape[0]
  *     cdef int node_id = 0             # <<<<<<<<<<<<<<
- *     cdef int K = values.shape[1]
- *     for i in xrange(n):
+ *     cdef int n_outputs = values.shape[1]
+ *     cdef int n_classes = values.shape[2]
  */
   __pyx_v_node_id = 0;
 
-  /* "sklearn/tree/_tree.pyx":473
+  /* "sklearn/tree/_tree.pyx":668
  *     cdef int n = X.shape[0]
  *     cdef int node_id = 0
- *     cdef int K = values.shape[1]             # <<<<<<<<<<<<<<
- *     for i in xrange(n):
- *         node_id = 0
+ *     cdef int n_outputs = values.shape[1]             # <<<<<<<<<<<<<<
+ *     cdef int n_classes = values.shape[2]
+ * 
  */
-  __pyx_v_K = (__pyx_v_values->dimensions[1]);
+  __pyx_v_n_outputs = (__pyx_v_values->dimensions[1]);
 
-  /* "sklearn/tree/_tree.pyx":474
+  /* "sklearn/tree/_tree.pyx":669
  *     cdef int node_id = 0
- *     cdef int K = values.shape[1]
- *     for i in xrange(n):             # <<<<<<<<<<<<<<
+ *     cdef int n_outputs = values.shape[1]
+ *     cdef int n_classes = values.shape[2]             # <<<<<<<<<<<<<<
+ * 
+ *     for i from 0 <= i < n:
+ */
+  __pyx_v_n_classes = (__pyx_v_values->dimensions[2]);
+
+  /* "sklearn/tree/_tree.pyx":671
+ *     cdef int n_classes = values.shape[2]
+ * 
+ *     for i from 0 <= i < n:             # <<<<<<<<<<<<<<
  *         node_id = 0
  *         # While node_id not a leaf
  */
   __pyx_t_1 = __pyx_v_n;
-  for (__pyx_t_2 = 0; __pyx_t_2 < __pyx_t_1; __pyx_t_2+=1) {
-    __pyx_v_i = __pyx_t_2;
+  for (__pyx_v_i = 0; __pyx_v_i < __pyx_t_1; __pyx_v_i++) {
 
-    /* "sklearn/tree/_tree.pyx":475
- *     cdef int K = values.shape[1]
- *     for i in xrange(n):
+    /* "sklearn/tree/_tree.pyx":672
+ * 
+ *     for i from 0 <= i < n:
  *         node_id = 0             # <<<<<<<<<<<<<<
  *         # While node_id not a leaf
  *         while children[node_id, 0] != -1 and children[node_id, 1] != -1:
  */
     __pyx_v_node_id = 0;
 
-    /* "sklearn/tree/_tree.pyx":477
+    /* "sklearn/tree/_tree.pyx":674
  *         node_id = 0
  *         # While node_id not a leaf
  *         while children[node_id, 0] != -1 and children[node_id, 1] != -1:             # <<<<<<<<<<<<<<
@@ -4466,84 +5797,96 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_4_predict_tree(CYTHON_UNUSED PyO
  *                 node_id = children[node_id, 0]
  */
     while (1) {
-      __pyx_t_3 = __pyx_v_node_id;
-      __pyx_t_4 = 0;
-      __pyx_t_5 = ((*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_3, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_4, __pyx_pybuffernd_children.diminfo[1].strides)) != -1);
-      if (__pyx_t_5) {
-        __pyx_t_6 = __pyx_v_node_id;
-        __pyx_t_7 = 1;
-        __pyx_t_8 = ((*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_6, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_7, __pyx_pybuffernd_children.diminfo[1].strides)) != -1);
-        __pyx_t_9 = __pyx_t_8;
+      __pyx_t_2 = __pyx_v_node_id;
+      __pyx_t_3 = 0;
+      __pyx_t_4 = ((*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_2, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_3, __pyx_pybuffernd_children.diminfo[1].strides)) != -1);
+      if (__pyx_t_4) {
+        __pyx_t_5 = __pyx_v_node_id;
+        __pyx_t_6 = 1;
+        __pyx_t_7 = ((*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_5, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_6, __pyx_pybuffernd_children.diminfo[1].strides)) != -1);
+        __pyx_t_8 = __pyx_t_7;
       } else {
-        __pyx_t_9 = __pyx_t_5;
+        __pyx_t_8 = __pyx_t_4;
       }
-      if (!__pyx_t_9) break;
+      if (!__pyx_t_8) break;
 
-      /* "sklearn/tree/_tree.pyx":478
+      /* "sklearn/tree/_tree.pyx":675
  *         # While node_id not a leaf
  *         while children[node_id, 0] != -1 and children[node_id, 1] != -1:
  *             if X[i, feature[node_id]] <= threshold[node_id]:             # <<<<<<<<<<<<<<
  *                 node_id = children[node_id, 0]
  *             else:
  */
-      __pyx_t_10 = __pyx_v_node_id;
-      __pyx_t_11 = __pyx_v_i;
-      __pyx_t_12 = (*__Pyx_BufPtrStrided1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_feature.rcbuffer->pybuffer.buf, __pyx_t_10, __pyx_pybuffernd_feature.diminfo[0].strides));
-      __pyx_t_13 = __pyx_v_node_id;
-      __pyx_t_9 = ((*__Pyx_BufPtrStrided2d(__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_pybuffernd_X.rcbuffer->pybuffer.buf, __pyx_t_11, __pyx_pybuffernd_X.diminfo[0].strides, __pyx_t_12, __pyx_pybuffernd_X.diminfo[1].strides)) <= (*__Pyx_BufPtrStrided1d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_threshold.rcbuffer->pybuffer.buf, __pyx_t_13, __pyx_pybuffernd_threshold.diminfo[0].strides)));
-      if (__pyx_t_9) {
+      __pyx_t_9 = __pyx_v_node_id;
+      __pyx_t_10 = __pyx_v_i;
+      __pyx_t_11 = (*__Pyx_BufPtrStrided1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_feature.rcbuffer->pybuffer.buf, __pyx_t_9, __pyx_pybuffernd_feature.diminfo[0].strides));
+      __pyx_t_12 = __pyx_v_node_id;
+      __pyx_t_8 = ((*__Pyx_BufPtrStrided2d(__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_pybuffernd_X.rcbuffer->pybuffer.buf, __pyx_t_10, __pyx_pybuffernd_X.diminfo[0].strides, __pyx_t_11, __pyx_pybuffernd_X.diminfo[1].strides)) <= (*__Pyx_BufPtrStrided1d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_threshold.rcbuffer->pybuffer.buf, __pyx_t_12, __pyx_pybuffernd_threshold.diminfo[0].strides)));
+      if (__pyx_t_8) {
 
-        /* "sklearn/tree/_tree.pyx":479
+        /* "sklearn/tree/_tree.pyx":676
  *         while children[node_id, 0] != -1 and children[node_id, 1] != -1:
  *             if X[i, feature[node_id]] <= threshold[node_id]:
  *                 node_id = children[node_id, 0]             # <<<<<<<<<<<<<<
  *             else:
  *                 node_id = children[node_id, 1]
  */
-        __pyx_t_14 = __pyx_v_node_id;
-        __pyx_t_15 = 0;
-        __pyx_v_node_id = (*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_14, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_15, __pyx_pybuffernd_children.diminfo[1].strides));
+        __pyx_t_13 = __pyx_v_node_id;
+        __pyx_t_14 = 0;
+        __pyx_v_node_id = (*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_13, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_14, __pyx_pybuffernd_children.diminfo[1].strides));
         goto __pyx_L7;
       }
       /*else*/ {
 
-        /* "sklearn/tree/_tree.pyx":481
+        /* "sklearn/tree/_tree.pyx":678
  *                 node_id = children[node_id, 0]
  *             else:
  *                 node_id = children[node_id, 1]             # <<<<<<<<<<<<<<
- *         for k in xrange(K):
- *             pred[i, k] = values[node_id, k]
+ * 
+ *         for k from 0 <= k < n_outputs:
  */
-        __pyx_t_16 = __pyx_v_node_id;
-        __pyx_t_17 = 1;
-        __pyx_v_node_id = (*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_16, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_17, __pyx_pybuffernd_children.diminfo[1].strides));
+        __pyx_t_15 = __pyx_v_node_id;
+        __pyx_t_16 = 1;
+        __pyx_v_node_id = (*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_children.rcbuffer->pybuffer.buf, __pyx_t_15, __pyx_pybuffernd_children.diminfo[0].strides, __pyx_t_16, __pyx_pybuffernd_children.diminfo[1].strides));
       }
       __pyx_L7:;
     }
 
-    /* "sklearn/tree/_tree.pyx":482
- *             else:
+    /* "sklearn/tree/_tree.pyx":680
  *                 node_id = children[node_id, 1]
- *         for k in xrange(K):             # <<<<<<<<<<<<<<
- *             pred[i, k] = values[node_id, k]
  * 
+ *         for k from 0 <= k < n_outputs:             # <<<<<<<<<<<<<<
+ *             for c from 0 <= c < n_classes:
+ *                 pred[i, k, c] = values[node_id, k, c]
  */
-    __pyx_t_18 = __pyx_v_K;
-    for (__pyx_t_19 = 0; __pyx_t_19 < __pyx_t_18; __pyx_t_19+=1) {
-      __pyx_v_k = __pyx_t_19;
+    __pyx_t_17 = __pyx_v_n_outputs;
+    for (__pyx_v_k = 0; __pyx_v_k < __pyx_t_17; __pyx_v_k++) {
 
-      /* "sklearn/tree/_tree.pyx":483
- *                 node_id = children[node_id, 1]
- *         for k in xrange(K):
- *             pred[i, k] = values[node_id, k]             # <<<<<<<<<<<<<<
+      /* "sklearn/tree/_tree.pyx":681
+ * 
+ *         for k from 0 <= k < n_outputs:
+ *             for c from 0 <= c < n_classes:             # <<<<<<<<<<<<<<
+ *                 pred[i, k, c] = values[node_id, k, c]
+ * 
+ */
+      __pyx_t_18 = __pyx_v_n_classes;
+      for (__pyx_v_c = 0; __pyx_v_c < __pyx_t_18; __pyx_v_c++) {
+
+        /* "sklearn/tree/_tree.pyx":682
+ *         for k from 0 <= k < n_outputs:
+ *             for c from 0 <= c < n_classes:
+ *                 pred[i, k, c] = values[node_id, k, c]             # <<<<<<<<<<<<<<
  * 
  * 
  */
-      __pyx_t_20 = __pyx_v_node_id;
-      __pyx_t_21 = __pyx_v_k;
-      __pyx_t_22 = __pyx_v_i;
-      __pyx_t_23 = __pyx_v_k;
-      *__Pyx_BufPtrStrided2d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_pred.rcbuffer->pybuffer.buf, __pyx_t_22, __pyx_pybuffernd_pred.diminfo[0].strides, __pyx_t_23, __pyx_pybuffernd_pred.diminfo[1].strides) = (*__Pyx_BufPtrStrided2d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_values.rcbuffer->pybuffer.buf, __pyx_t_20, __pyx_pybuffernd_values.diminfo[0].strides, __pyx_t_21, __pyx_pybuffernd_values.diminfo[1].strides));
+        __pyx_t_19 = __pyx_v_node_id;
+        __pyx_t_20 = __pyx_v_k;
+        __pyx_t_21 = __pyx_v_c;
+        __pyx_t_22 = __pyx_v_i;
+        __pyx_t_23 = __pyx_v_k;
+        __pyx_t_24 = __pyx_v_c;
+        *__Pyx_BufPtrStrided3d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_pred.rcbuffer->pybuffer.buf, __pyx_t_22, __pyx_pybuffernd_pred.diminfo[0].strides, __pyx_t_23, __pyx_pybuffernd_pred.diminfo[1].strides, __pyx_t_24, __pyx_pybuffernd_pred.diminfo[2].strides) = (*__Pyx_BufPtrStrided3d(__pyx_t_5numpy_float64_t *, __pyx_pybuffernd_values.rcbuffer->pybuffer.buf, __pyx_t_19, __pyx_pybuffernd_values.diminfo[0].strides, __pyx_t_20, __pyx_pybuffernd_values.diminfo[1].strides, __pyx_t_21, __pyx_pybuffernd_values.diminfo[2].strides));
+      }
     }
   }
 
@@ -4612,23 +5955,23 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_7_error_at_leaf(PyObject *__pyx_
         values[1] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__sample_mask);
         if (likely(values[1])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_error_at_leaf", 1, 4, 4, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_error_at_leaf", 1, 4, 4, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  2:
         values[2] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__criterion);
         if (likely(values[2])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_error_at_leaf", 1, 4, 4, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_error_at_leaf", 1, 4, 4, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  3:
         values[3] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__n_samples);
         if (likely(values[3])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_error_at_leaf", 1, 4, 4, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_error_at_leaf", 1, 4, 4, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
       }
       if (unlikely(kw_args > 0)) {
-        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_error_at_leaf") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_error_at_leaf") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
       }
     } else if (PyTuple_GET_SIZE(__pyx_args) != 4) {
       goto __pyx_L5_argtuple_error;
@@ -4641,19 +5984,19 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_7_error_at_leaf(PyObject *__pyx_
     __pyx_v_y = ((PyArrayObject *)values[0]);
     __pyx_v_sample_mask = ((PyArrayObject *)values[1]);
     __pyx_v_criterion = ((struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *)values[2]);
-    __pyx_v_n_samples = __Pyx_PyInt_AsInt(values[3]); if (unlikely((__pyx_v_n_samples == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 488; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_n_samples = __Pyx_PyInt_AsInt(values[3]); if (unlikely((__pyx_v_n_samples == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 688; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
   }
   goto __pyx_L4_argument_unpacking_done;
   __pyx_L5_argtuple_error:;
-  __Pyx_RaiseArgtupleInvalid("_error_at_leaf", 1, 4, 4, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+  __Pyx_RaiseArgtupleInvalid("_error_at_leaf", 1, 4, 4, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
   __pyx_L3_error:;
   __Pyx_AddTraceback("sklearn.tree._tree._error_at_leaf", __pyx_clineno, __pyx_lineno, __pyx_filename);
   __Pyx_RefNannyFinishContext();
   return NULL;
   __pyx_L4_argument_unpacking_done:;
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_y), __pyx_ptype_5numpy_ndarray, 1, "y", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_sample_mask), __pyx_ptype_5numpy_ndarray, 1, "sample_mask", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 487; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_criterion), __pyx_ptype_7sklearn_4tree_5_tree_Criterion, 1, "criterion", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 487; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_y), __pyx_ptype_5numpy_ndarray, 1, "y", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_sample_mask), __pyx_ptype_5numpy_ndarray, 1, "sample_mask", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 686; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_criterion), __pyx_ptype_7sklearn_4tree_5_tree_Criterion, 1, "criterion", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 687; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_6_error_at_leaf(__pyx_self, __pyx_v_y, __pyx_v_sample_mask, __pyx_v_criterion, __pyx_v_n_samples);
   goto __pyx_L0;
   __pyx_L1_error:;
@@ -4663,17 +6006,18 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_7_error_at_leaf(PyObject *__pyx_
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":486
+/* "sklearn/tree/_tree.pyx":685
  * 
  * 
- * def _error_at_leaf(np.ndarray[DTYPE_t, ndim=1, mode="c"] y,             # <<<<<<<<<<<<<<
- *                    np.ndarray sample_mask, Criterion criterion,
- *                    int n_samples):
+ * def _error_at_leaf(np.ndarray[DTYPE_t, ndim=2, mode="c"] y,             # <<<<<<<<<<<<<<
+ *                    np.ndarray sample_mask,
+ *                    Criterion criterion,
  */
 
 static PyObject *__pyx_pf_7sklearn_4tree_5_tree_6_error_at_leaf(CYTHON_UNUSED PyObject *__pyx_self, PyArrayObject *__pyx_v_y, PyArrayObject *__pyx_v_sample_mask, struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *__pyx_v_criterion, int __pyx_v_n_samples) {
-  int __pyx_v_n_total_samples;
   __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y_ptr;
+  int __pyx_v_y_stride;
+  int __pyx_v_n_total_samples;
   __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask_ptr;
   __Pyx_LocalBuf_ND __pyx_pybuffernd_y;
   __Pyx_Buffer __pyx_pybuffer_y;
@@ -4690,55 +6034,64 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_6_error_at_leaf(CYTHON_UNUSED Py
   __pyx_pybuffernd_y.rcbuffer = &__pyx_pybuffer_y;
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_y.rcbuffer->pybuffer, (PyObject*)__pyx_v_y, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_y.rcbuffer->pybuffer, (PyObject*)__pyx_v_y, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
-  __pyx_pybuffernd_y.diminfo[0].strides = __pyx_pybuffernd_y.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_y.diminfo[0].shape = __pyx_pybuffernd_y.rcbuffer->pybuffer.shape[0];
+  __pyx_pybuffernd_y.diminfo[0].strides = __pyx_pybuffernd_y.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_y.diminfo[0].shape = __pyx_pybuffernd_y.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_y.diminfo[1].strides = __pyx_pybuffernd_y.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_y.diminfo[1].shape = __pyx_pybuffernd_y.rcbuffer->pybuffer.shape[1];
 
-  /* "sklearn/tree/_tree.pyx":491
+  /* "sklearn/tree/_tree.pyx":691
  *     """Compute criterion error at leaf with terminal region defined
  *     by `sample_mask`. """
- *     cdef int n_total_samples = y.shape[0]             # <<<<<<<<<<<<<<
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data             # <<<<<<<<<<<<<<
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
+ *     cdef int n_total_samples = y.shape[0]
  */
-  __pyx_v_n_total_samples = (__pyx_v_y->dimensions[0]);
+  __pyx_v_y_ptr = ((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *)__pyx_v_y->data);
 
-  /* "sklearn/tree/_tree.pyx":492
+  /* "sklearn/tree/_tree.pyx":692
  *     by `sample_mask`. """
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]             # <<<<<<<<<<<<<<
  *     cdef int n_total_samples = y.shape[0]
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data             # <<<<<<<<<<<<<<
  *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
  */
-  __pyx_v_y_ptr = ((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *)__pyx_v_y->data);
+  __pyx_v_y_stride = (((int)(__pyx_v_y->strides[0])) / ((int)(__pyx_v_y->strides[1])));
 
-  /* "sklearn/tree/_tree.pyx":493
+  /* "sklearn/tree/_tree.pyx":693
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
+ *     cdef int n_total_samples = y.shape[0]             # <<<<<<<<<<<<<<
+ *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
+ * 
+ */
+  __pyx_v_n_total_samples = (__pyx_v_y->dimensions[0]);
+
+  /* "sklearn/tree/_tree.pyx":694
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
  *     cdef int n_total_samples = y.shape[0]
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
  *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data             # <<<<<<<<<<<<<<
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
- *     return criterion.eval()
+ * 
+ *     criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)
  */
   __pyx_v_sample_mask_ptr = ((__pyx_t_7sklearn_4tree_5_tree_BOOL_t *)__pyx_v_sample_mask->data);
 
-  /* "sklearn/tree/_tree.pyx":494
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
+  /* "sklearn/tree/_tree.pyx":696
  *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)             # <<<<<<<<<<<<<<
- *     return criterion.eval()
  * 
+ *     criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)             # <<<<<<<<<<<<<<
+ * 
+ *     return criterion.eval()
  */
-  ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->init(__pyx_v_criterion, __pyx_v_y_ptr, __pyx_v_sample_mask_ptr, __pyx_v_n_samples, __pyx_v_n_total_samples);
+  ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->init(__pyx_v_criterion, __pyx_v_y_ptr, __pyx_v_y_stride, __pyx_v_sample_mask_ptr, __pyx_v_n_samples, __pyx_v_n_total_samples);
 
-  /* "sklearn/tree/_tree.pyx":495
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
+  /* "sklearn/tree/_tree.pyx":698
+ *     criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)
+ * 
  *     return criterion.eval()             # <<<<<<<<<<<<<<
  * 
  * 
  */
   __Pyx_XDECREF(__pyx_r);
-  __pyx_t_1 = PyFloat_FromDouble(((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->eval(__pyx_v_criterion)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 495; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = PyFloat_FromDouble(((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->eval(__pyx_v_criterion)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 698; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_1);
   __pyx_r = __pyx_t_1;
   __pyx_t_1 = 0;
@@ -4763,7 +6116,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_6_error_at_leaf(CYTHON_UNUSED Py
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":498
+/* "sklearn/tree/_tree.pyx":701
  * 
  * 
  * cdef int smallest_sample_larger_than(int sample_idx, DTYPE_t *X_i,             # <<<<<<<<<<<<<<
@@ -4781,7 +6134,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
   int __pyx_t_2;
   __Pyx_RefNannySetupContext("smallest_sample_larger_than", 0);
 
-  /* "sklearn/tree/_tree.pyx":515
+  /* "sklearn/tree/_tree.pyx":718
  *         -1 if no such element exists.
  *     """
  *     cdef int idx = 0, j             # <<<<<<<<<<<<<<
@@ -4790,7 +6143,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
  */
   __pyx_v_idx = 0;
 
-  /* "sklearn/tree/_tree.pyx":516
+  /* "sklearn/tree/_tree.pyx":719
  *     """
  *     cdef int idx = 0, j
  *     cdef DTYPE_t threshold = -DBL_MAX             # <<<<<<<<<<<<<<
@@ -4799,7 +6152,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
  */
   __pyx_v_threshold = (-DBL_MAX);
 
-  /* "sklearn/tree/_tree.pyx":518
+  /* "sklearn/tree/_tree.pyx":721
  *     cdef DTYPE_t threshold = -DBL_MAX
  * 
  *     if sample_idx > -1:             # <<<<<<<<<<<<<<
@@ -4809,7 +6162,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
   __pyx_t_1 = (__pyx_v_sample_idx > -1);
   if (__pyx_t_1) {
 
-    /* "sklearn/tree/_tree.pyx":519
+    /* "sklearn/tree/_tree.pyx":722
  * 
  *     if sample_idx > -1:
  *         threshold = X_i[X_argsorted_i[sample_idx]]             # <<<<<<<<<<<<<<
@@ -4821,7 +6174,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
   }
   __pyx_L3:;
 
-  /* "sklearn/tree/_tree.pyx":521
+  /* "sklearn/tree/_tree.pyx":724
  *         threshold = X_i[X_argsorted_i[sample_idx]]
  * 
  *     for idx from sample_idx < idx < n_total_samples:             # <<<<<<<<<<<<<<
@@ -4831,7 +6184,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
   __pyx_t_2 = __pyx_v_n_total_samples;
   for (__pyx_v_idx = __pyx_v_sample_idx+1; __pyx_v_idx < __pyx_t_2; __pyx_v_idx++) {
 
-    /* "sklearn/tree/_tree.pyx":522
+    /* "sklearn/tree/_tree.pyx":725
  * 
  *     for idx from sample_idx < idx < n_total_samples:
  *         j = X_argsorted_i[idx]             # <<<<<<<<<<<<<<
@@ -4840,7 +6193,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
  */
     __pyx_v_j = (__pyx_v_X_argsorted_i[__pyx_v_idx]);
 
-    /* "sklearn/tree/_tree.pyx":524
+    /* "sklearn/tree/_tree.pyx":727
  *         j = X_argsorted_i[idx]
  * 
  *         if sample_mask[j] == 0:             # <<<<<<<<<<<<<<
@@ -4850,7 +6203,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
     __pyx_t_1 = ((__pyx_v_sample_mask[__pyx_v_j]) == 0);
     if (__pyx_t_1) {
 
-      /* "sklearn/tree/_tree.pyx":525
+      /* "sklearn/tree/_tree.pyx":728
  * 
  *         if sample_mask[j] == 0:
  *             continue             # <<<<<<<<<<<<<<
@@ -4862,7 +6215,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
     }
     __pyx_L6:;
 
-    /* "sklearn/tree/_tree.pyx":527
+    /* "sklearn/tree/_tree.pyx":730
  *             continue
  * 
  *         if X_i[j] > threshold + 1.e-7:             # <<<<<<<<<<<<<<
@@ -4872,7 +6225,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
     __pyx_t_1 = ((__pyx_v_X_i[__pyx_v_j]) > (__pyx_v_threshold + 1.e-7));
     if (__pyx_t_1) {
 
-      /* "sklearn/tree/_tree.pyx":528
+      /* "sklearn/tree/_tree.pyx":731
  * 
  *         if X_i[j] > threshold + 1.e-7:
  *             return idx             # <<<<<<<<<<<<<<
@@ -4887,7 +6240,7 @@ static int __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(int __pyx_v
     __pyx_L4_continue:;
   }
 
-  /* "sklearn/tree/_tree.pyx":530
+  /* "sklearn/tree/_tree.pyx":733
  *             return idx
  * 
  *     return -1             # <<<<<<<<<<<<<<
@@ -4950,53 +6303,53 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_9_find_best_split(PyObject *__py
         values[1] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__y);
         if (likely(values[1])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  2:
         values[2] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__X_argsorted);
         if (likely(values[2])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  3:
         values[3] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__sample_mask);
         if (likely(values[3])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  4:
         values[4] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__n_samples);
         if (likely(values[4])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 4); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 4); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  5:
         values[5] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__min_leaf);
         if (likely(values[5])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 5); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 5); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  6:
         values[6] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__max_features);
         if (likely(values[6])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 6); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 6); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  7:
         values[7] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__criterion);
         if (likely(values[7])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 7); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 7); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  8:
         values[8] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__random_state);
         if (likely(values[8])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 8); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, 8); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
       }
       if (unlikely(kw_args > 0)) {
-        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_find_best_split") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_find_best_split") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
       }
     } else if (PyTuple_GET_SIZE(__pyx_args) != 9) {
       goto __pyx_L5_argtuple_error;
@@ -5015,25 +6368,25 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_9_find_best_split(PyObject *__py
     __pyx_v_y = ((PyArrayObject *)values[1]);
     __pyx_v_X_argsorted = ((PyArrayObject *)values[2]);
     __pyx_v_sample_mask = ((PyArrayObject *)values[3]);
-    __pyx_v_n_samples = __Pyx_PyInt_AsInt(values[4]); if (unlikely((__pyx_v_n_samples == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 537; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
-    __pyx_v_min_leaf = __Pyx_PyInt_AsInt(values[5]); if (unlikely((__pyx_v_min_leaf == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 538; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
-    __pyx_v_max_features = __Pyx_PyInt_AsInt(values[6]); if (unlikely((__pyx_v_max_features == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 539; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_n_samples = __Pyx_PyInt_AsInt(values[4]); if (unlikely((__pyx_v_n_samples == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 740; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_min_leaf = __Pyx_PyInt_AsInt(values[5]); if (unlikely((__pyx_v_min_leaf == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 741; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_max_features = __Pyx_PyInt_AsInt(values[6]); if (unlikely((__pyx_v_max_features == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 742; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
     __pyx_v_criterion = ((struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *)values[7]);
     __pyx_v_random_state = values[8];
   }
   goto __pyx_L4_argument_unpacking_done;
   __pyx_L5_argtuple_error:;
-  __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+  __Pyx_RaiseArgtupleInvalid("_find_best_split", 1, 9, 9, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
   __pyx_L3_error:;
   __Pyx_AddTraceback("sklearn.tree._tree._find_best_split", __pyx_clineno, __pyx_lineno, __pyx_filename);
   __Pyx_RefNannyFinishContext();
   return NULL;
   __pyx_L4_argument_unpacking_done:;
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X), __pyx_ptype_5numpy_ndarray, 1, "X", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_y), __pyx_ptype_5numpy_ndarray, 1, "y", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 534; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X_argsorted), __pyx_ptype_5numpy_ndarray, 1, "X_argsorted", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 535; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_sample_mask), __pyx_ptype_5numpy_ndarray, 1, "sample_mask", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 536; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_criterion), __pyx_ptype_7sklearn_4tree_5_tree_Criterion, 1, "criterion", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 540; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X), __pyx_ptype_5numpy_ndarray, 1, "X", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_y), __pyx_ptype_5numpy_ndarray, 1, "y", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 737; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X_argsorted), __pyx_ptype_5numpy_ndarray, 1, "X_argsorted", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 738; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_sample_mask), __pyx_ptype_5numpy_ndarray, 1, "sample_mask", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 739; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_criterion), __pyx_ptype_7sklearn_4tree_5_tree_Criterion, 1, "criterion", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 743; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(__pyx_self, __pyx_v_X, __pyx_v_y, __pyx_v_X_argsorted, __pyx_v_sample_mask, __pyx_v_n_samples, __pyx_v_min_leaf, __pyx_v_max_features, __pyx_v_criterion, __pyx_v_random_state);
   goto __pyx_L0;
   __pyx_L1_error:;
@@ -5043,11 +6396,11 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_9_find_best_split(PyObject *__py
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":533
+/* "sklearn/tree/_tree.pyx":736
  * 
  * 
  * def _find_best_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,             # <<<<<<<<<<<<<<
- *                      np.ndarray[DTYPE_t, ndim=1, mode="c"] y,
+ *                      np.ndarray[DTYPE_t, ndim=2, mode="c"] y,
  *                      np.ndarray[np.int32_t, ndim=2, mode="fortran"] X_argsorted,
  */
 
@@ -5065,11 +6418,12 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
   __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_v_error;
   __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_v_best_error;
   __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_v_best_t;
-  __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y_ptr;
   __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_X_i;
   int *__pyx_v_X_argsorted_i;
+  __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y_ptr;
   __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask_ptr;
   PyArrayObject *__pyx_v_features = 0;
+  int __pyx_v_y_stride;
   int __pyx_v_X_elem_stride;
   int __pyx_v_X_col_stride;
   int __pyx_v_X_stride;
@@ -5086,22 +6440,20 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
   __Pyx_Buffer __pyx_pybuffer_y;
   PyObject *__pyx_r = NULL;
   __Pyx_RefNannyDeclarations
-  PyObject *__pyx_t_1 = NULL;
-  PyObject *__pyx_t_2 = NULL;
-  __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_t_3;
-  PyArrayObject *__pyx_t_4 = NULL;
-  int __pyx_t_5;
+  PyArrayObject *__pyx_t_1 = NULL;
+  int __pyx_t_2;
+  PyObject *__pyx_t_3 = NULL;
+  PyObject *__pyx_t_4 = NULL;
+  PyObject *__pyx_t_5 = NULL;
   PyObject *__pyx_t_6 = NULL;
   PyObject *__pyx_t_7 = NULL;
-  PyObject *__pyx_t_8 = NULL;
-  int __pyx_t_9;
+  int __pyx_t_8;
+  PyObject *__pyx_t_9 = NULL;
   PyObject *__pyx_t_10 = NULL;
   PyObject *__pyx_t_11 = NULL;
-  PyObject *__pyx_t_12 = NULL;
+  int __pyx_t_12;
   int __pyx_t_13;
-  int __pyx_t_14;
-  __pyx_t_5numpy_int32_t __pyx_t_15;
-  __pyx_t_5numpy_int32_t __pyx_t_16;
+  __pyx_t_5numpy_int32_t __pyx_t_14;
   int __pyx_lineno = 0;
   const char *__pyx_filename = NULL;
   int __pyx_clineno = 0;
@@ -5124,21 +6476,21 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
   __pyx_pybuffernd_X_argsorted.rcbuffer = &__pyx_pybuffer_X_argsorted;
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X.rcbuffer->pybuffer, (PyObject*)__pyx_v_X, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_F_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X.rcbuffer->pybuffer, (PyObject*)__pyx_v_X, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_F_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_X.diminfo[0].strides = __pyx_pybuffernd_X.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_X.diminfo[0].shape = __pyx_pybuffernd_X.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_X.diminfo[1].strides = __pyx_pybuffernd_X.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_X.diminfo[1].shape = __pyx_pybuffernd_X.rcbuffer->pybuffer.shape[1];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_y.rcbuffer->pybuffer, (PyObject*)__pyx_v_y, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_y.rcbuffer->pybuffer, (PyObject*)__pyx_v_y, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
-  __pyx_pybuffernd_y.diminfo[0].strides = __pyx_pybuffernd_y.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_y.diminfo[0].shape = __pyx_pybuffernd_y.rcbuffer->pybuffer.shape[0];
+  __pyx_pybuffernd_y.diminfo[0].strides = __pyx_pybuffernd_y.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_y.diminfo[0].shape = __pyx_pybuffernd_y.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_y.diminfo[1].strides = __pyx_pybuffernd_y.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_y.diminfo[1].shape = __pyx_pybuffernd_y.rcbuffer->pybuffer.shape[1];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer, (PyObject*)__pyx_v_X_argsorted, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_F_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer, (PyObject*)__pyx_v_X_argsorted, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_F_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_X_argsorted.diminfo[0].strides = __pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_X_argsorted.diminfo[0].shape = __pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_X_argsorted.diminfo[1].strides = __pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_X_argsorted.diminfo[1].shape = __pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer.shape[1];
 
-  /* "sklearn/tree/_tree.pyx":592
+  /* "sklearn/tree/_tree.pyx":795
  *     """
  *     # Variables declarations
  *     cdef int n_total_samples = X.shape[0]             # <<<<<<<<<<<<<<
@@ -5147,7 +6499,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_n_total_samples = (__pyx_v_X->dimensions[0]);
 
-  /* "sklearn/tree/_tree.pyx":593
+  /* "sklearn/tree/_tree.pyx":796
  *     # Variables declarations
  *     cdef int n_total_samples = X.shape[0]
  *     cdef int n_features = X.shape[1]             # <<<<<<<<<<<<<<
@@ -5156,7 +6508,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_n_features = (__pyx_v_X->dimensions[1]);
 
-  /* "sklearn/tree/_tree.pyx":594
+  /* "sklearn/tree/_tree.pyx":797
  *     cdef int n_total_samples = X.shape[0]
  *     cdef int n_features = X.shape[1]
  *     cdef int i, a, b, best_i = -1             # <<<<<<<<<<<<<<
@@ -5165,7 +6517,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_best_i = -1;
 
-  /* "sklearn/tree/_tree.pyx":595
+  /* "sklearn/tree/_tree.pyx":798
  *     cdef int n_features = X.shape[1]
  *     cdef int i, a, b, best_i = -1
  *     cdef np.int32_t feature_idx = -1             # <<<<<<<<<<<<<<
@@ -5174,106 +6526,101 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_feature_idx = -1;
 
-  /* "sklearn/tree/_tree.pyx":596
+  /* "sklearn/tree/_tree.pyx":799
  *     cdef int i, a, b, best_i = -1
  *     cdef np.int32_t feature_idx = -1
  *     cdef int n_left = 0             # <<<<<<<<<<<<<<
  *     cdef DTYPE_t t, initial_error, error
- *     cdef DTYPE_t best_error = np.inf, best_t = np.inf
+ *     cdef DTYPE_t best_error = INFINITY, best_t = INFINITY
  */
   __pyx_v_n_left = 0;
 
-  /* "sklearn/tree/_tree.pyx":598
+  /* "sklearn/tree/_tree.pyx":801
  *     cdef int n_left = 0
  *     cdef DTYPE_t t, initial_error, error
- *     cdef DTYPE_t best_error = np.inf, best_t = np.inf             # <<<<<<<<<<<<<<
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
- *     cdef DTYPE_t *X_i = NULL
+ *     cdef DTYPE_t best_error = INFINITY, best_t = INFINITY             # <<<<<<<<<<<<<<
+ *     cdef DTYPE_t* X_i = NULL
+ *     cdef int* X_argsorted_i = NULL
  */
-  __pyx_t_1 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 598; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_2 = PyObject_GetAttr(__pyx_t_1, __pyx_n_s__inf); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 598; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __pyx_t_3 = __pyx_PyFloat_AsFloat(__pyx_t_2); if (unlikely((__pyx_t_3 == (npy_float32)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 598; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  __pyx_v_best_error = __pyx_t_3;
-  __pyx_t_2 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 598; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __pyx_t_1 = PyObject_GetAttr(__pyx_t_2, __pyx_n_s__inf); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 598; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  __pyx_t_3 = __pyx_PyFloat_AsFloat(__pyx_t_1); if (unlikely((__pyx_t_3 == (npy_float32)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 598; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __pyx_v_best_t = __pyx_t_3;
+  __pyx_v_best_error = __pyx_v_7sklearn_4tree_5_tree_INFINITY;
+  __pyx_v_best_t = __pyx_v_7sklearn_4tree_5_tree_INFINITY;
 
-  /* "sklearn/tree/_tree.pyx":599
+  /* "sklearn/tree/_tree.pyx":802
  *     cdef DTYPE_t t, initial_error, error
- *     cdef DTYPE_t best_error = np.inf, best_t = np.inf
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data             # <<<<<<<<<<<<<<
- *     cdef DTYPE_t *X_i = NULL
- *     cdef int *X_argsorted_i = NULL
- */
-  __pyx_v_y_ptr = ((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *)__pyx_v_y->data);
-
-  /* "sklearn/tree/_tree.pyx":600
- *     cdef DTYPE_t best_error = np.inf, best_t = np.inf
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
- *     cdef DTYPE_t *X_i = NULL             # <<<<<<<<<<<<<<
- *     cdef int *X_argsorted_i = NULL
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
+ *     cdef DTYPE_t best_error = INFINITY, best_t = INFINITY
+ *     cdef DTYPE_t* X_i = NULL             # <<<<<<<<<<<<<<
+ *     cdef int* X_argsorted_i = NULL
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
  */
   __pyx_v_X_i = NULL;
 
-  /* "sklearn/tree/_tree.pyx":601
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
- *     cdef DTYPE_t *X_i = NULL
- *     cdef int *X_argsorted_i = NULL             # <<<<<<<<<<<<<<
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
- *     cdef np.ndarray[np.int32_t, ndim=1, mode='c'] features = None
+  /* "sklearn/tree/_tree.pyx":803
+ *     cdef DTYPE_t best_error = INFINITY, best_t = INFINITY
+ *     cdef DTYPE_t* X_i = NULL
+ *     cdef int* X_argsorted_i = NULL             # <<<<<<<<<<<<<<
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+ *     cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data
  */
   __pyx_v_X_argsorted_i = NULL;
 
-  /* "sklearn/tree/_tree.pyx":602
- *     cdef DTYPE_t *X_i = NULL
- *     cdef int *X_argsorted_i = NULL
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data             # <<<<<<<<<<<<<<
- *     cdef np.ndarray[np.int32_t, ndim=1, mode='c'] features = None
+  /* "sklearn/tree/_tree.pyx":804
+ *     cdef DTYPE_t* X_i = NULL
+ *     cdef int* X_argsorted_i = NULL
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data             # <<<<<<<<<<<<<<
+ *     cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data
+ *     cdef np.ndarray[np.int32_t, ndim=1, mode="c"] features = None
+ */
+  __pyx_v_y_ptr = ((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *)__pyx_v_y->data);
+
+  /* "sklearn/tree/_tree.pyx":805
+ *     cdef int* X_argsorted_i = NULL
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+ *     cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data             # <<<<<<<<<<<<<<
+ *     cdef np.ndarray[np.int32_t, ndim=1, mode="c"] features = None
  * 
  */
   __pyx_v_sample_mask_ptr = ((__pyx_t_7sklearn_4tree_5_tree_BOOL_t *)__pyx_v_sample_mask->data);
 
-  /* "sklearn/tree/_tree.pyx":603
- *     cdef int *X_argsorted_i = NULL
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
- *     cdef np.ndarray[np.int32_t, ndim=1, mode='c'] features = None             # <<<<<<<<<<<<<<
+  /* "sklearn/tree/_tree.pyx":806
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+ *     cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data
+ *     cdef np.ndarray[np.int32_t, ndim=1, mode="c"] features = None             # <<<<<<<<<<<<<<
  * 
  *     # Compute the column strides (increment in pointer elements to get
  */
-  __pyx_t_4 = ((PyArrayObject *)Py_None);
+  __pyx_t_1 = ((PyArrayObject *)Py_None);
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_4, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_1, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {
       __pyx_v_features = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None); __pyx_pybuffernd_features.rcbuffer->pybuffer.buf = NULL;
-      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 603; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 806; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     } else {__pyx_pybuffernd_features.diminfo[0].strides = __pyx_pybuffernd_features.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_features.diminfo[0].shape = __pyx_pybuffernd_features.rcbuffer->pybuffer.shape[0];
     }
   }
-  __pyx_t_4 = 0;
+  __pyx_t_1 = 0;
   __Pyx_INCREF(Py_None);
   __pyx_v_features = ((PyArrayObject *)Py_None);
 
-  /* "sklearn/tree/_tree.pyx":607
+  /* "sklearn/tree/_tree.pyx":810
  *     # Compute the column strides (increment in pointer elements to get
  *     # from column i to i + 1) for `X` and `X_argsorted`
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]             # <<<<<<<<<<<<<<
+ *     cdef int X_elem_stride = X.strides[0]
+ *     cdef int X_col_stride = X.strides[1]
+ */
+  __pyx_v_y_stride = (((int)(__pyx_v_y->strides[0])) / ((int)(__pyx_v_y->strides[1])));
+
+  /* "sklearn/tree/_tree.pyx":811
+ *     # from column i to i + 1) for `X` and `X_argsorted`
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
  *     cdef int X_elem_stride = X.strides[0]             # <<<<<<<<<<<<<<
  *     cdef int X_col_stride = X.strides[1]
  *     cdef int X_stride = X_col_stride / X_elem_stride
  */
   __pyx_v_X_elem_stride = (__pyx_v_X->strides[0]);
 
-  /* "sklearn/tree/_tree.pyx":608
- *     # from column i to i + 1) for `X` and `X_argsorted`
+  /* "sklearn/tree/_tree.pyx":812
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
  *     cdef int X_elem_stride = X.strides[0]
  *     cdef int X_col_stride = X.strides[1]             # <<<<<<<<<<<<<<
  *     cdef int X_stride = X_col_stride / X_elem_stride
@@ -5281,7 +6628,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_X_col_stride = (__pyx_v_X->strides[1]);
 
-  /* "sklearn/tree/_tree.pyx":609
+  /* "sklearn/tree/_tree.pyx":813
  *     cdef int X_elem_stride = X.strides[0]
  *     cdef int X_col_stride = X.strides[1]
  *     cdef int X_stride = X_col_stride / X_elem_stride             # <<<<<<<<<<<<<<
@@ -5290,7 +6637,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_X_stride = (__pyx_v_X_col_stride / __pyx_v_X_elem_stride);
 
-  /* "sklearn/tree/_tree.pyx":610
+  /* "sklearn/tree/_tree.pyx":814
  *     cdef int X_col_stride = X.strides[1]
  *     cdef int X_stride = X_col_stride / X_elem_stride
  *     cdef int X_argsorted_elem_stride = X_argsorted.strides[0]             # <<<<<<<<<<<<<<
@@ -5299,7 +6646,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_X_argsorted_elem_stride = (__pyx_v_X_argsorted->strides[0]);
 
-  /* "sklearn/tree/_tree.pyx":611
+  /* "sklearn/tree/_tree.pyx":815
  *     cdef int X_stride = X_col_stride / X_elem_stride
  *     cdef int X_argsorted_elem_stride = X_argsorted.strides[0]
  *     cdef int X_argsorted_col_stride = X_argsorted.strides[1]             # <<<<<<<<<<<<<<
@@ -5308,7 +6655,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_X_argsorted_col_stride = (__pyx_v_X_argsorted->strides[1]);
 
-  /* "sklearn/tree/_tree.pyx":612
+  /* "sklearn/tree/_tree.pyx":816
  *     cdef int X_argsorted_elem_stride = X_argsorted.strides[0]
  *     cdef int X_argsorted_col_stride = X_argsorted.strides[1]
  *     cdef int X_argsorted_stride = X_argsorted_col_stride / X_argsorted_elem_stride             # <<<<<<<<<<<<<<
@@ -5317,44 +6664,44 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_X_argsorted_stride = (__pyx_v_X_argsorted_col_stride / __pyx_v_X_argsorted_elem_stride);
 
-  /* "sklearn/tree/_tree.pyx":615
+  /* "sklearn/tree/_tree.pyx":819
  * 
  *     # Compute the initial criterion value in the node
- *     X_argsorted_i = <int *>X_argsorted.data             # <<<<<<<<<<<<<<
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
+ *     X_argsorted_i = <int*>X_argsorted.data             # <<<<<<<<<<<<<<
+ *     criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)
  *     initial_error = criterion.eval()
  */
   __pyx_v_X_argsorted_i = ((int *)__pyx_v_X_argsorted->data);
 
-  /* "sklearn/tree/_tree.pyx":616
+  /* "sklearn/tree/_tree.pyx":820
  *     # Compute the initial criterion value in the node
- *     X_argsorted_i = <int *>X_argsorted.data
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)             # <<<<<<<<<<<<<<
+ *     X_argsorted_i = <int*>X_argsorted.data
+ *     criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)             # <<<<<<<<<<<<<<
  *     initial_error = criterion.eval()
  * 
  */
-  ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->init(__pyx_v_criterion, __pyx_v_y_ptr, __pyx_v_sample_mask_ptr, __pyx_v_n_samples, __pyx_v_n_total_samples);
+  ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->init(__pyx_v_criterion, __pyx_v_y_ptr, __pyx_v_y_stride, __pyx_v_sample_mask_ptr, __pyx_v_n_samples, __pyx_v_n_total_samples);
 
-  /* "sklearn/tree/_tree.pyx":617
- *     X_argsorted_i = <int *>X_argsorted.data
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
+  /* "sklearn/tree/_tree.pyx":821
+ *     X_argsorted_i = <int*>X_argsorted.data
+ *     criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)
  *     initial_error = criterion.eval()             # <<<<<<<<<<<<<<
  * 
  *     if initial_error == 0:  # break early if the node is pure
  */
   __pyx_v_initial_error = ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->eval(__pyx_v_criterion);
 
-  /* "sklearn/tree/_tree.pyx":619
+  /* "sklearn/tree/_tree.pyx":823
  *     initial_error = criterion.eval()
  * 
  *     if initial_error == 0:  # break early if the node is pure             # <<<<<<<<<<<<<<
  *         return best_i, best_t, initial_error, initial_error
  * 
  */
-  __pyx_t_5 = (__pyx_v_initial_error == 0.0);
-  if (__pyx_t_5) {
+  __pyx_t_2 = (__pyx_v_initial_error == 0.0);
+  if (__pyx_t_2) {
 
-    /* "sklearn/tree/_tree.pyx":620
+    /* "sklearn/tree/_tree.pyx":824
  * 
  *     if initial_error == 0:  # break early if the node is pure
  *         return best_i, best_t, initial_error, initial_error             # <<<<<<<<<<<<<<
@@ -5362,36 +6709,36 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  *     best_error = initial_error
  */
     __Pyx_XDECREF(__pyx_r);
-    __pyx_t_1 = PyInt_FromLong(__pyx_v_best_i); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 620; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_1);
-    __pyx_t_2 = PyFloat_FromDouble(__pyx_v_best_t); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 620; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_2);
-    __pyx_t_6 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 620; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_3 = PyInt_FromLong(__pyx_v_best_i); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 824; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_3);
+    __pyx_t_4 = PyFloat_FromDouble(__pyx_v_best_t); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 824; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_4);
+    __pyx_t_5 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 824; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_5);
+    __pyx_t_6 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 824; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     __Pyx_GOTREF(__pyx_t_6);
-    __pyx_t_7 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 620; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_7 = PyTuple_New(4); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 824; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     __Pyx_GOTREF(__pyx_t_7);
-    __pyx_t_8 = PyTuple_New(4); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 620; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_8);
-    PyTuple_SET_ITEM(__pyx_t_8, 0, __pyx_t_1);
-    __Pyx_GIVEREF(__pyx_t_1);
-    PyTuple_SET_ITEM(__pyx_t_8, 1, __pyx_t_2);
-    __Pyx_GIVEREF(__pyx_t_2);
-    PyTuple_SET_ITEM(__pyx_t_8, 2, __pyx_t_6);
+    PyTuple_SET_ITEM(__pyx_t_7, 0, __pyx_t_3);
+    __Pyx_GIVEREF(__pyx_t_3);
+    PyTuple_SET_ITEM(__pyx_t_7, 1, __pyx_t_4);
+    __Pyx_GIVEREF(__pyx_t_4);
+    PyTuple_SET_ITEM(__pyx_t_7, 2, __pyx_t_5);
+    __Pyx_GIVEREF(__pyx_t_5);
+    PyTuple_SET_ITEM(__pyx_t_7, 3, __pyx_t_6);
     __Pyx_GIVEREF(__pyx_t_6);
-    PyTuple_SET_ITEM(__pyx_t_8, 3, __pyx_t_7);
-    __Pyx_GIVEREF(__pyx_t_7);
-    __pyx_t_1 = 0;
-    __pyx_t_2 = 0;
+    __pyx_t_3 = 0;
+    __pyx_t_4 = 0;
+    __pyx_t_5 = 0;
     __pyx_t_6 = 0;
+    __pyx_r = ((PyObject *)__pyx_t_7);
     __pyx_t_7 = 0;
-    __pyx_r = ((PyObject *)__pyx_t_8);
-    __pyx_t_8 = 0;
     goto __pyx_L0;
     goto __pyx_L3;
   }
   __pyx_L3:;
 
-  /* "sklearn/tree/_tree.pyx":622
+  /* "sklearn/tree/_tree.pyx":826
  *         return best_i, best_t, initial_error, initial_error
  * 
  *     best_error = initial_error             # <<<<<<<<<<<<<<
@@ -5400,79 +6747,79 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
   __pyx_v_best_error = __pyx_v_initial_error;
 
-  /* "sklearn/tree/_tree.pyx":625
+  /* "sklearn/tree/_tree.pyx":829
  * 
  *     # Features to consider
  *     features = np.arange(n_features, dtype=np.int32)             # <<<<<<<<<<<<<<
  *     if max_features < 0 or max_features >= n_features:
  *         max_features = n_features
  */
-  __pyx_t_8 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_8);
-  __pyx_t_7 = PyObject_GetAttr(__pyx_t_8, __pyx_n_s__arange); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_7 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_7);
-  __Pyx_DECREF(__pyx_t_8); __pyx_t_8 = 0;
-  __pyx_t_8 = PyInt_FromLong(__pyx_v_n_features); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_8);
-  __pyx_t_6 = PyTuple_New(1); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_6 = PyObject_GetAttr(__pyx_t_7, __pyx_n_s__arange); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_6);
-  PyTuple_SET_ITEM(__pyx_t_6, 0, __pyx_t_8);
-  __Pyx_GIVEREF(__pyx_t_8);
-  __pyx_t_8 = 0;
-  __pyx_t_8 = PyDict_New(); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(((PyObject *)__pyx_t_8));
-  __pyx_t_2 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __pyx_t_1 = PyObject_GetAttr(__pyx_t_2, __pyx_n_s__int32); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  if (PyDict_SetItem(__pyx_t_8, ((PyObject *)__pyx_n_s__dtype), __pyx_t_1) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __pyx_t_1 = PyObject_Call(__pyx_t_7, ((PyObject *)__pyx_t_6), ((PyObject *)__pyx_t_8)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
   __Pyx_DECREF(__pyx_t_7); __pyx_t_7 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_6)); __pyx_t_6 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_8)); __pyx_t_8 = 0;
-  if (!(likely(((__pyx_t_1) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_1, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __pyx_t_4 = ((PyArrayObject *)__pyx_t_1);
+  __pyx_t_7 = PyInt_FromLong(__pyx_v_n_features); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_7);
+  __pyx_t_5 = PyTuple_New(1); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_5);
+  PyTuple_SET_ITEM(__pyx_t_5, 0, __pyx_t_7);
+  __Pyx_GIVEREF(__pyx_t_7);
+  __pyx_t_7 = 0;
+  __pyx_t_7 = PyDict_New(); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(((PyObject *)__pyx_t_7));
+  __pyx_t_4 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_4);
+  __pyx_t_3 = PyObject_GetAttr(__pyx_t_4, __pyx_n_s__int32); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_3);
+  __Pyx_DECREF(__pyx_t_4); __pyx_t_4 = 0;
+  if (PyDict_SetItem(__pyx_t_7, ((PyObject *)__pyx_n_s__dtype), __pyx_t_3) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_3); __pyx_t_3 = 0;
+  __pyx_t_3 = PyObject_Call(__pyx_t_6, ((PyObject *)__pyx_t_5), ((PyObject *)__pyx_t_7)); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_3);
+  __Pyx_DECREF(__pyx_t_6); __pyx_t_6 = 0;
+  __Pyx_DECREF(((PyObject *)__pyx_t_5)); __pyx_t_5 = 0;
+  __Pyx_DECREF(((PyObject *)__pyx_t_7)); __pyx_t_7 = 0;
+  if (!(likely(((__pyx_t_3) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_3, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = ((PyArrayObject *)__pyx_t_3);
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
     __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_features.rcbuffer->pybuffer);
-    __pyx_t_9 = __Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_4, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack);
-    if (unlikely(__pyx_t_9 < 0)) {
-      PyErr_Fetch(&__pyx_t_10, &__pyx_t_11, &__pyx_t_12);
+    __pyx_t_8 = __Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_1, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack);
+    if (unlikely(__pyx_t_8 < 0)) {
+      PyErr_Fetch(&__pyx_t_9, &__pyx_t_10, &__pyx_t_11);
       if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_v_features, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {
-        Py_XDECREF(__pyx_t_10); Py_XDECREF(__pyx_t_11); Py_XDECREF(__pyx_t_12);
+        Py_XDECREF(__pyx_t_9); Py_XDECREF(__pyx_t_10); Py_XDECREF(__pyx_t_11);
         __Pyx_RaiseBufferFallbackError();
       } else {
-        PyErr_Restore(__pyx_t_10, __pyx_t_11, __pyx_t_12);
+        PyErr_Restore(__pyx_t_9, __pyx_t_10, __pyx_t_11);
       }
     }
     __pyx_pybuffernd_features.diminfo[0].strides = __pyx_pybuffernd_features.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_features.diminfo[0].shape = __pyx_pybuffernd_features.rcbuffer->pybuffer.shape[0];
-    if (unlikely(__pyx_t_9 < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 625; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__pyx_t_8 < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 829; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
-  __pyx_t_4 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_v_features));
-  __pyx_v_features = ((PyArrayObject *)__pyx_t_1);
   __pyx_t_1 = 0;
+  __Pyx_DECREF(((PyObject *)__pyx_v_features));
+  __pyx_v_features = ((PyArrayObject *)__pyx_t_3);
+  __pyx_t_3 = 0;
 
-  /* "sklearn/tree/_tree.pyx":626
+  /* "sklearn/tree/_tree.pyx":830
  *     # Features to consider
  *     features = np.arange(n_features, dtype=np.int32)
  *     if max_features < 0 or max_features >= n_features:             # <<<<<<<<<<<<<<
  *         max_features = n_features
  *     else:
  */
-  __pyx_t_5 = (__pyx_v_max_features < 0);
-  if (!__pyx_t_5) {
-    __pyx_t_13 = (__pyx_v_max_features >= __pyx_v_n_features);
-    __pyx_t_14 = __pyx_t_13;
+  __pyx_t_2 = (__pyx_v_max_features < 0);
+  if (!__pyx_t_2) {
+    __pyx_t_12 = (__pyx_v_max_features >= __pyx_v_n_features);
+    __pyx_t_13 = __pyx_t_12;
   } else {
-    __pyx_t_14 = __pyx_t_5;
+    __pyx_t_13 = __pyx_t_2;
   }
-  if (__pyx_t_14) {
+  if (__pyx_t_13) {
 
-    /* "sklearn/tree/_tree.pyx":627
+    /* "sklearn/tree/_tree.pyx":831
  *     features = np.arange(n_features, dtype=np.int32)
  *     if max_features < 0 or max_features >= n_features:
  *         max_features = n_features             # <<<<<<<<<<<<<<
@@ -5484,92 +6831,91 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
   }
   /*else*/ {
 
-    /* "sklearn/tree/_tree.pyx":629
+    /* "sklearn/tree/_tree.pyx":833
  *         max_features = n_features
  *     else:
  *         features = random_state.permutation(features)[:max_features]             # <<<<<<<<<<<<<<
  * 
  *     # Look for the best split
  */
-    __pyx_t_1 = PyObject_GetAttr(__pyx_v_random_state, __pyx_n_s__permutation); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 629; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_1);
-    __pyx_t_8 = PyTuple_New(1); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 629; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_8);
+    __pyx_t_3 = PyObject_GetAttr(__pyx_v_random_state, __pyx_n_s__permutation); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 833; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_3);
+    __pyx_t_7 = PyTuple_New(1); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 833; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_7);
     __Pyx_INCREF(((PyObject *)__pyx_v_features));
-    PyTuple_SET_ITEM(__pyx_t_8, 0, ((PyObject *)__pyx_v_features));
+    PyTuple_SET_ITEM(__pyx_t_7, 0, ((PyObject *)__pyx_v_features));
     __Pyx_GIVEREF(((PyObject *)__pyx_v_features));
-    __pyx_t_6 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_t_8), NULL); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 629; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_6);
-    __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-    __Pyx_DECREF(((PyObject *)__pyx_t_8)); __pyx_t_8 = 0;
-    __pyx_t_8 = __Pyx_PySequence_GetSlice(__pyx_t_6, 0, __pyx_v_max_features); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 629; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_8);
-    __Pyx_DECREF(__pyx_t_6); __pyx_t_6 = 0;
-    if (!(likely(((__pyx_t_8) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_8, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 629; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __pyx_t_4 = ((PyArrayObject *)__pyx_t_8);
+    __pyx_t_5 = PyObject_Call(__pyx_t_3, ((PyObject *)__pyx_t_7), NULL); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 833; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_5);
+    __Pyx_DECREF(__pyx_t_3); __pyx_t_3 = 0;
+    __Pyx_DECREF(((PyObject *)__pyx_t_7)); __pyx_t_7 = 0;
+    __pyx_t_7 = __Pyx_PySequence_GetSlice(__pyx_t_5, 0, __pyx_v_max_features); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 833; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_7);
+    __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
+    if (!(likely(((__pyx_t_7) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_7, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 833; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_1 = ((PyArrayObject *)__pyx_t_7);
     {
       __Pyx_BufFmt_StackElem __pyx_stack[1];
       __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_features.rcbuffer->pybuffer);
-      __pyx_t_9 = __Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_4, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack);
-      if (unlikely(__pyx_t_9 < 0)) {
-        PyErr_Fetch(&__pyx_t_12, &__pyx_t_11, &__pyx_t_10);
+      __pyx_t_8 = __Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_1, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack);
+      if (unlikely(__pyx_t_8 < 0)) {
+        PyErr_Fetch(&__pyx_t_11, &__pyx_t_10, &__pyx_t_9);
         if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_v_features, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {
-          Py_XDECREF(__pyx_t_12); Py_XDECREF(__pyx_t_11); Py_XDECREF(__pyx_t_10);
+          Py_XDECREF(__pyx_t_11); Py_XDECREF(__pyx_t_10); Py_XDECREF(__pyx_t_9);
           __Pyx_RaiseBufferFallbackError();
         } else {
-          PyErr_Restore(__pyx_t_12, __pyx_t_11, __pyx_t_10);
+          PyErr_Restore(__pyx_t_11, __pyx_t_10, __pyx_t_9);
         }
       }
       __pyx_pybuffernd_features.diminfo[0].strides = __pyx_pybuffernd_features.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_features.diminfo[0].shape = __pyx_pybuffernd_features.rcbuffer->pybuffer.shape[0];
-      if (unlikely(__pyx_t_9 < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 629; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      if (unlikely(__pyx_t_8 < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 833; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     }
-    __pyx_t_4 = 0;
+    __pyx_t_1 = 0;
     __Pyx_DECREF(((PyObject *)__pyx_v_features));
-    __pyx_v_features = ((PyArrayObject *)__pyx_t_8);
-    __pyx_t_8 = 0;
+    __pyx_v_features = ((PyArrayObject *)__pyx_t_7);
+    __pyx_t_7 = 0;
   }
   __pyx_L4:;
 
-  /* "sklearn/tree/_tree.pyx":632
+  /* "sklearn/tree/_tree.pyx":836
  * 
  *     # Look for the best split
- *     for feature_idx in range(max_features):             # <<<<<<<<<<<<<<
+ *     for feature_idx from 0 <= feature_idx < max_features:             # <<<<<<<<<<<<<<
  *         i = features[feature_idx]
  *         # Get i-th col of X and X_sorted
  */
-  __pyx_t_9 = __pyx_v_max_features;
-  for (__pyx_t_15 = 0; __pyx_t_15 < __pyx_t_9; __pyx_t_15+=1) {
-    __pyx_v_feature_idx = __pyx_t_15;
+  __pyx_t_8 = __pyx_v_max_features;
+  for (__pyx_v_feature_idx = 0; __pyx_v_feature_idx < __pyx_t_8; __pyx_v_feature_idx++) {
 
-    /* "sklearn/tree/_tree.pyx":633
+    /* "sklearn/tree/_tree.pyx":837
  *     # Look for the best split
- *     for feature_idx in range(max_features):
+ *     for feature_idx from 0 <= feature_idx < max_features:
  *         i = features[feature_idx]             # <<<<<<<<<<<<<<
  *         # Get i-th col of X and X_sorted
- *         X_i = (<DTYPE_t *>X.data) + X_stride * i
+ *         X_i = (<DTYPE_t*>X.data) + X_stride * i
  */
-    __pyx_t_16 = __pyx_v_feature_idx;
-    __pyx_v_i = (*__Pyx_BufPtrCContig1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_features.rcbuffer->pybuffer.buf, __pyx_t_16, __pyx_pybuffernd_features.diminfo[0].strides));
+    __pyx_t_14 = __pyx_v_feature_idx;
+    __pyx_v_i = (*__Pyx_BufPtrCContig1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_features.rcbuffer->pybuffer.buf, __pyx_t_14, __pyx_pybuffernd_features.diminfo[0].strides));
 
-    /* "sklearn/tree/_tree.pyx":635
+    /* "sklearn/tree/_tree.pyx":839
  *         i = features[feature_idx]
  *         # Get i-th col of X and X_sorted
- *         X_i = (<DTYPE_t *>X.data) + X_stride * i             # <<<<<<<<<<<<<<
- *         X_argsorted_i = (<int *>X_argsorted.data) + X_argsorted_stride * i
+ *         X_i = (<DTYPE_t*>X.data) + X_stride * i             # <<<<<<<<<<<<<<
+ *         X_argsorted_i = (<int*>X_argsorted.data) + X_argsorted_stride * i
  * 
  */
     __pyx_v_X_i = (((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *)__pyx_v_X->data) + (__pyx_v_X_stride * __pyx_v_i));
 
-    /* "sklearn/tree/_tree.pyx":636
+    /* "sklearn/tree/_tree.pyx":840
  *         # Get i-th col of X and X_sorted
- *         X_i = (<DTYPE_t *>X.data) + X_stride * i
- *         X_argsorted_i = (<int *>X_argsorted.data) + X_argsorted_stride * i             # <<<<<<<<<<<<<<
+ *         X_i = (<DTYPE_t*>X.data) + X_stride * i
+ *         X_argsorted_i = (<int*>X_argsorted.data) + X_argsorted_stride * i             # <<<<<<<<<<<<<<
  * 
  *         # Reset the criterion for this feature
  */
     __pyx_v_X_argsorted_i = (((int *)__pyx_v_X_argsorted->data) + (__pyx_v_X_argsorted_stride * __pyx_v_i));
 
-    /* "sklearn/tree/_tree.pyx":639
+    /* "sklearn/tree/_tree.pyx":843
  * 
  *         # Reset the criterion for this feature
  *         criterion.reset()             # <<<<<<<<<<<<<<
@@ -5578,7 +6924,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
     ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->reset(__pyx_v_criterion);
 
-    /* "sklearn/tree/_tree.pyx":642
+    /* "sklearn/tree/_tree.pyx":846
  * 
  *         # Index of smallest sample in X_argsorted_i that is in the sample mask
  *         a = 0             # <<<<<<<<<<<<<<
@@ -5587,7 +6933,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
     __pyx_v_a = 0;
 
-    /* "sklearn/tree/_tree.pyx":643
+    /* "sklearn/tree/_tree.pyx":847
  *         # Index of smallest sample in X_argsorted_i that is in the sample mask
  *         a = 0
  *         while sample_mask_ptr[X_argsorted_i[a]] == 0:             # <<<<<<<<<<<<<<
@@ -5595,10 +6941,10 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  * 
  */
     while (1) {
-      __pyx_t_14 = ((__pyx_v_sample_mask_ptr[(__pyx_v_X_argsorted_i[__pyx_v_a])]) == 0);
-      if (!__pyx_t_14) break;
+      __pyx_t_13 = ((__pyx_v_sample_mask_ptr[(__pyx_v_X_argsorted_i[__pyx_v_a])]) == 0);
+      if (!__pyx_t_13) break;
 
-      /* "sklearn/tree/_tree.pyx":644
+      /* "sklearn/tree/_tree.pyx":848
  *         a = 0
  *         while sample_mask_ptr[X_argsorted_i[a]] == 0:
  *             a = a + 1             # <<<<<<<<<<<<<<
@@ -5608,7 +6954,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
       __pyx_v_a = (__pyx_v_a + 1);
     }
 
-    /* "sklearn/tree/_tree.pyx":647
+    /* "sklearn/tree/_tree.pyx":851
  * 
  *         # Consider splits between two consecutive samples
  *         while True:             # <<<<<<<<<<<<<<
@@ -5618,7 +6964,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
     while (1) {
       if (!1) break;
 
-      /* "sklearn/tree/_tree.pyx":650
+      /* "sklearn/tree/_tree.pyx":854
  *             # Find the following larger sample
  *             b = smallest_sample_larger_than(a, X_i, X_argsorted_i,
  *                                             sample_mask_ptr, n_total_samples)             # <<<<<<<<<<<<<<
@@ -5627,17 +6973,17 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
       __pyx_v_b = __pyx_f_7sklearn_4tree_5_tree_smallest_sample_larger_than(__pyx_v_a, __pyx_v_X_i, __pyx_v_X_argsorted_i, __pyx_v_sample_mask_ptr, __pyx_v_n_total_samples);
 
-      /* "sklearn/tree/_tree.pyx":651
+      /* "sklearn/tree/_tree.pyx":855
  *             b = smallest_sample_larger_than(a, X_i, X_argsorted_i,
  *                                             sample_mask_ptr, n_total_samples)
  *             if b == -1:             # <<<<<<<<<<<<<<
  *                 break
  * 
  */
-      __pyx_t_14 = (__pyx_v_b == -1);
-      if (__pyx_t_14) {
+      __pyx_t_13 = (__pyx_v_b == -1);
+      if (__pyx_t_13) {
 
-        /* "sklearn/tree/_tree.pyx":652
+        /* "sklearn/tree/_tree.pyx":856
  *                                             sample_mask_ptr, n_total_samples)
  *             if b == -1:
  *                 break             # <<<<<<<<<<<<<<
@@ -5649,32 +6995,32 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
       }
       __pyx_L11:;
 
-      /* "sklearn/tree/_tree.pyx":655
+      /* "sklearn/tree/_tree.pyx":859
  * 
  *             # Better split than the best so far?
- *             n_left = criterion.update(a, b, y_ptr, X_argsorted_i, sample_mask_ptr)             # <<<<<<<<<<<<<<
+ *             n_left = criterion.update(a, b, y_ptr, y_stride, X_argsorted_i, sample_mask_ptr)             # <<<<<<<<<<<<<<
  * 
  *             # Only consider splits that respect min_leaf
  */
-      __pyx_v_n_left = ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->update(__pyx_v_criterion, __pyx_v_a, __pyx_v_b, __pyx_v_y_ptr, __pyx_v_X_argsorted_i, __pyx_v_sample_mask_ptr);
+      __pyx_v_n_left = ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->update(__pyx_v_criterion, __pyx_v_a, __pyx_v_b, __pyx_v_y_ptr, __pyx_v_y_stride, __pyx_v_X_argsorted_i, __pyx_v_sample_mask_ptr);
 
-      /* "sklearn/tree/_tree.pyx":658
+      /* "sklearn/tree/_tree.pyx":862
  * 
  *             # Only consider splits that respect min_leaf
  *             if n_left < min_leaf or (n_samples - n_left) < min_leaf:             # <<<<<<<<<<<<<<
  *                 a = b
  *                 continue
  */
-      __pyx_t_14 = (__pyx_v_n_left < __pyx_v_min_leaf);
-      if (!__pyx_t_14) {
-        __pyx_t_5 = ((__pyx_v_n_samples - __pyx_v_n_left) < __pyx_v_min_leaf);
-        __pyx_t_13 = __pyx_t_5;
+      __pyx_t_13 = (__pyx_v_n_left < __pyx_v_min_leaf);
+      if (!__pyx_t_13) {
+        __pyx_t_2 = ((__pyx_v_n_samples - __pyx_v_n_left) < __pyx_v_min_leaf);
+        __pyx_t_12 = __pyx_t_2;
       } else {
-        __pyx_t_13 = __pyx_t_14;
+        __pyx_t_12 = __pyx_t_13;
       }
-      if (__pyx_t_13) {
+      if (__pyx_t_12) {
 
-        /* "sklearn/tree/_tree.pyx":659
+        /* "sklearn/tree/_tree.pyx":863
  *             # Only consider splits that respect min_leaf
  *             if n_left < min_leaf or (n_samples - n_left) < min_leaf:
  *                 a = b             # <<<<<<<<<<<<<<
@@ -5683,7 +7029,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
         __pyx_v_a = __pyx_v_b;
 
-        /* "sklearn/tree/_tree.pyx":660
+        /* "sklearn/tree/_tree.pyx":864
  *             if n_left < min_leaf or (n_samples - n_left) < min_leaf:
  *                 a = b
  *                 continue             # <<<<<<<<<<<<<<
@@ -5695,7 +7041,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
       }
       __pyx_L12:;
 
-      /* "sklearn/tree/_tree.pyx":662
+      /* "sklearn/tree/_tree.pyx":866
  *                 continue
  * 
  *             error = criterion.eval()             # <<<<<<<<<<<<<<
@@ -5704,17 +7050,17 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
       __pyx_v_error = ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->eval(__pyx_v_criterion);
 
-      /* "sklearn/tree/_tree.pyx":664
+      /* "sklearn/tree/_tree.pyx":868
  *             error = criterion.eval()
  * 
  *             if error < best_error:             # <<<<<<<<<<<<<<
  *                 t = X_i[X_argsorted_i[a]] + \
  *                     ((X_i[X_argsorted_i[b]] - X_i[X_argsorted_i[a]]) / 2.0)
  */
-      __pyx_t_13 = (__pyx_v_error < __pyx_v_best_error);
-      if (__pyx_t_13) {
+      __pyx_t_12 = (__pyx_v_error < __pyx_v_best_error);
+      if (__pyx_t_12) {
 
-        /* "sklearn/tree/_tree.pyx":666
+        /* "sklearn/tree/_tree.pyx":870
  *             if error < best_error:
  *                 t = X_i[X_argsorted_i[a]] + \
  *                     ((X_i[X_argsorted_i[b]] - X_i[X_argsorted_i[a]]) / 2.0)             # <<<<<<<<<<<<<<
@@ -5723,17 +7069,17 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
         __pyx_v_t = ((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_a])]) + (((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_b])]) - (__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_a])])) / 2.0));
 
-        /* "sklearn/tree/_tree.pyx":667
+        /* "sklearn/tree/_tree.pyx":871
  *                 t = X_i[X_argsorted_i[a]] + \
  *                     ((X_i[X_argsorted_i[b]] - X_i[X_argsorted_i[a]]) / 2.0)
  *                 if t == X_i[X_argsorted_i[b]]:             # <<<<<<<<<<<<<<
  *                     t = X_i[X_argsorted_i[a]]
  *                 best_i = i
  */
-        __pyx_t_13 = (__pyx_v_t == (__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_b])]));
-        if (__pyx_t_13) {
+        __pyx_t_12 = (__pyx_v_t == (__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_b])]));
+        if (__pyx_t_12) {
 
-          /* "sklearn/tree/_tree.pyx":668
+          /* "sklearn/tree/_tree.pyx":872
  *                     ((X_i[X_argsorted_i[b]] - X_i[X_argsorted_i[a]]) / 2.0)
  *                 if t == X_i[X_argsorted_i[b]]:
  *                     t = X_i[X_argsorted_i[a]]             # <<<<<<<<<<<<<<
@@ -5745,7 +7091,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
         }
         __pyx_L14:;
 
-        /* "sklearn/tree/_tree.pyx":669
+        /* "sklearn/tree/_tree.pyx":873
  *                 if t == X_i[X_argsorted_i[b]]:
  *                     t = X_i[X_argsorted_i[a]]
  *                 best_i = i             # <<<<<<<<<<<<<<
@@ -5754,7 +7100,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
         __pyx_v_best_i = __pyx_v_i;
 
-        /* "sklearn/tree/_tree.pyx":670
+        /* "sklearn/tree/_tree.pyx":874
  *                     t = X_i[X_argsorted_i[a]]
  *                 best_i = i
  *                 best_t = t             # <<<<<<<<<<<<<<
@@ -5763,7 +7109,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  */
         __pyx_v_best_t = __pyx_v_t;
 
-        /* "sklearn/tree/_tree.pyx":671
+        /* "sklearn/tree/_tree.pyx":875
  *                 best_i = i
  *                 best_t = t
  *                 best_error = error             # <<<<<<<<<<<<<<
@@ -5775,7 +7121,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
       }
       __pyx_L13:;
 
-      /* "sklearn/tree/_tree.pyx":674
+      /* "sklearn/tree/_tree.pyx":878
  * 
  *             # Proceed to the next interval
  *             a = b             # <<<<<<<<<<<<<<
@@ -5788,7 +7134,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
     __pyx_L10_break:;
   }
 
-  /* "sklearn/tree/_tree.pyx":676
+  /* "sklearn/tree/_tree.pyx":880
  *             a = b
  * 
  *     return best_i, best_t, best_error, initial_error             # <<<<<<<<<<<<<<
@@ -5796,40 +7142,40 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_8_find_best_split(CYTHON_UNUSED
  * def _find_best_random_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
  */
   __Pyx_XDECREF(__pyx_r);
-  __pyx_t_8 = PyInt_FromLong(__pyx_v_best_i); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 676; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_8);
-  __pyx_t_6 = PyFloat_FromDouble(__pyx_v_best_t); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 676; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_6);
-  __pyx_t_1 = PyFloat_FromDouble(__pyx_v_best_error); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 676; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_7 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 676; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_7 = PyInt_FromLong(__pyx_v_best_i); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 880; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_7);
-  __pyx_t_2 = PyTuple_New(4); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 676; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  PyTuple_SET_ITEM(__pyx_t_2, 0, __pyx_t_8);
-  __Pyx_GIVEREF(__pyx_t_8);
-  PyTuple_SET_ITEM(__pyx_t_2, 1, __pyx_t_6);
-  __Pyx_GIVEREF(__pyx_t_6);
-  PyTuple_SET_ITEM(__pyx_t_2, 2, __pyx_t_1);
-  __Pyx_GIVEREF(__pyx_t_1);
-  PyTuple_SET_ITEM(__pyx_t_2, 3, __pyx_t_7);
+  __pyx_t_5 = PyFloat_FromDouble(__pyx_v_best_t); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 880; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_5);
+  __pyx_t_3 = PyFloat_FromDouble(__pyx_v_best_error); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 880; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_3);
+  __pyx_t_6 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 880; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_6);
+  __pyx_t_4 = PyTuple_New(4); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 880; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_4);
+  PyTuple_SET_ITEM(__pyx_t_4, 0, __pyx_t_7);
   __Pyx_GIVEREF(__pyx_t_7);
-  __pyx_t_8 = 0;
-  __pyx_t_6 = 0;
-  __pyx_t_1 = 0;
+  PyTuple_SET_ITEM(__pyx_t_4, 1, __pyx_t_5);
+  __Pyx_GIVEREF(__pyx_t_5);
+  PyTuple_SET_ITEM(__pyx_t_4, 2, __pyx_t_3);
+  __Pyx_GIVEREF(__pyx_t_3);
+  PyTuple_SET_ITEM(__pyx_t_4, 3, __pyx_t_6);
+  __Pyx_GIVEREF(__pyx_t_6);
   __pyx_t_7 = 0;
-  __pyx_r = ((PyObject *)__pyx_t_2);
-  __pyx_t_2 = 0;
+  __pyx_t_5 = 0;
+  __pyx_t_3 = 0;
+  __pyx_t_6 = 0;
+  __pyx_r = ((PyObject *)__pyx_t_4);
+  __pyx_t_4 = 0;
   goto __pyx_L0;
 
   __pyx_r = Py_None; __Pyx_INCREF(Py_None);
   goto __pyx_L0;
   __pyx_L1_error:;
-  __Pyx_XDECREF(__pyx_t_1);
-  __Pyx_XDECREF(__pyx_t_2);
+  __Pyx_XDECREF(__pyx_t_3);
+  __Pyx_XDECREF(__pyx_t_4);
+  __Pyx_XDECREF(__pyx_t_5);
   __Pyx_XDECREF(__pyx_t_6);
   __Pyx_XDECREF(__pyx_t_7);
-  __Pyx_XDECREF(__pyx_t_8);
   { PyObject *__pyx_type, *__pyx_value, *__pyx_tb;
     __Pyx_ErrFetch(&__pyx_type, &__pyx_value, &__pyx_tb);
     __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_X.rcbuffer->pybuffer);
@@ -5899,53 +7245,53 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_11_find_best_random_split(PyObje
         values[1] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__y);
         if (likely(values[1])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 1); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  2:
         values[2] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__X_argsorted);
         if (likely(values[2])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 2); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  3:
         values[3] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__sample_mask);
         if (likely(values[3])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 3); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  4:
         values[4] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__n_samples);
         if (likely(values[4])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 4); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 4); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  5:
         values[5] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__min_leaf);
         if (likely(values[5])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 5); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 5); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  6:
         values[6] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__max_features);
         if (likely(values[6])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 6); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 6); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  7:
         values[7] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__criterion);
         if (likely(values[7])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 7); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 7); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
         case  8:
         values[8] = PyDict_GetItem(__pyx_kwds, __pyx_n_s__random_state);
         if (likely(values[8])) kw_args--;
         else {
-          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 8); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+          __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, 8); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
         }
       }
       if (unlikely(kw_args > 0)) {
-        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_find_best_random_split") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+        if (unlikely(__Pyx_ParseOptionalKeywords(__pyx_kwds, __pyx_pyargnames, 0, values, pos_args, "_find_best_random_split") < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
       }
     } else if (PyTuple_GET_SIZE(__pyx_args) != 9) {
       goto __pyx_L5_argtuple_error;
@@ -5964,25 +7310,25 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_11_find_best_random_split(PyObje
     __pyx_v_y = ((PyArrayObject *)values[1]);
     __pyx_v_X_argsorted = ((PyArrayObject *)values[2]);
     __pyx_v_sample_mask = ((PyArrayObject *)values[3]);
-    __pyx_v_n_samples = __Pyx_PyInt_AsInt(values[4]); if (unlikely((__pyx_v_n_samples == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 682; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
-    __pyx_v_min_leaf = __Pyx_PyInt_AsInt(values[5]); if (unlikely((__pyx_v_min_leaf == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 683; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
-    __pyx_v_max_features = __Pyx_PyInt_AsInt(values[6]); if (unlikely((__pyx_v_max_features == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 684; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_n_samples = __Pyx_PyInt_AsInt(values[4]); if (unlikely((__pyx_v_n_samples == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 886; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_min_leaf = __Pyx_PyInt_AsInt(values[5]); if (unlikely((__pyx_v_min_leaf == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 887; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+    __pyx_v_max_features = __Pyx_PyInt_AsInt(values[6]); if (unlikely((__pyx_v_max_features == (int)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 888; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
     __pyx_v_criterion = ((struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *)values[7]);
     __pyx_v_random_state = values[8];
   }
   goto __pyx_L4_argument_unpacking_done;
   __pyx_L5_argtuple_error:;
-  __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
+  __Pyx_RaiseArgtupleInvalid("_find_best_random_split", 1, 9, 9, PyTuple_GET_SIZE(__pyx_args)); {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L3_error;}
   __pyx_L3_error:;
   __Pyx_AddTraceback("sklearn.tree._tree._find_best_random_split", __pyx_clineno, __pyx_lineno, __pyx_filename);
   __Pyx_RefNannyFinishContext();
   return NULL;
   __pyx_L4_argument_unpacking_done:;
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X), __pyx_ptype_5numpy_ndarray, 1, "X", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_y), __pyx_ptype_5numpy_ndarray, 1, "y", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 679; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X_argsorted), __pyx_ptype_5numpy_ndarray, 1, "X_argsorted", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 680; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_sample_mask), __pyx_ptype_5numpy_ndarray, 1, "sample_mask", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 681; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_criterion), __pyx_ptype_7sklearn_4tree_5_tree_Criterion, 1, "criterion", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X), __pyx_ptype_5numpy_ndarray, 1, "X", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_y), __pyx_ptype_5numpy_ndarray, 1, "y", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 883; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_X_argsorted), __pyx_ptype_5numpy_ndarray, 1, "X_argsorted", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 884; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_sample_mask), __pyx_ptype_5numpy_ndarray, 1, "sample_mask", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 885; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (unlikely(!__Pyx_ArgTypeTest(((PyObject *)__pyx_v_criterion), __pyx_ptype_7sklearn_4tree_5_tree_Criterion, 1, "criterion", 0))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 889; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_r = __pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(__pyx_self, __pyx_v_X, __pyx_v_y, __pyx_v_X_argsorted, __pyx_v_sample_mask, __pyx_v_n_samples, __pyx_v_min_leaf, __pyx_v_max_features, __pyx_v_criterion, __pyx_v_random_state);
   goto __pyx_L0;
   __pyx_L1_error:;
@@ -5992,11 +7338,11 @@ static PyObject *__pyx_pw_7sklearn_4tree_5_tree_11_find_best_random_split(PyObje
   return __pyx_r;
 }
 
-/* "sklearn/tree/_tree.pyx":678
+/* "sklearn/tree/_tree.pyx":882
  *     return best_i, best_t, best_error, initial_error
  * 
  * def _find_best_random_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,             # <<<<<<<<<<<<<<
- *                             np.ndarray[DTYPE_t, ndim=1, mode="c"] y,
+ *                             np.ndarray[DTYPE_t, ndim=2, mode="c"] y,
  *                             np.ndarray[np.int32_t, ndim=2, mode="fortran"] X_argsorted,
  */
 
@@ -6015,11 +7361,12 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
   __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_v_error;
   __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_v_best_error;
   __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_v_best_t;
-  __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y_ptr;
   __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_X_i;
   int *__pyx_v_X_argsorted_i;
+  __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *__pyx_v_y_ptr;
   __pyx_t_7sklearn_4tree_5_tree_BOOL_t *__pyx_v_sample_mask_ptr;
   PyArrayObject *__pyx_v_features = 0;
+  int __pyx_v_y_stride;
   int __pyx_v_X_elem_stride;
   int __pyx_v_X_col_stride;
   int __pyx_v_X_stride;
@@ -6036,22 +7383,21 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
   __Pyx_Buffer __pyx_pybuffer_y;
   PyObject *__pyx_r = NULL;
   __Pyx_RefNannyDeclarations
-  PyObject *__pyx_t_1 = NULL;
-  PyObject *__pyx_t_2 = NULL;
-  __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_t_3;
-  PyArrayObject *__pyx_t_4 = NULL;
-  int __pyx_t_5;
+  PyArrayObject *__pyx_t_1 = NULL;
+  int __pyx_t_2;
+  PyObject *__pyx_t_3 = NULL;
+  PyObject *__pyx_t_4 = NULL;
+  PyObject *__pyx_t_5 = NULL;
   PyObject *__pyx_t_6 = NULL;
   PyObject *__pyx_t_7 = NULL;
-  PyObject *__pyx_t_8 = NULL;
-  int __pyx_t_9;
+  int __pyx_t_8;
+  PyObject *__pyx_t_9 = NULL;
   PyObject *__pyx_t_10 = NULL;
   PyObject *__pyx_t_11 = NULL;
-  PyObject *__pyx_t_12 = NULL;
+  int __pyx_t_12;
   int __pyx_t_13;
-  int __pyx_t_14;
-  __pyx_t_5numpy_int32_t __pyx_t_15;
-  __pyx_t_5numpy_int32_t __pyx_t_16;
+  __pyx_t_5numpy_int32_t __pyx_t_14;
+  __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_t_15;
   int __pyx_lineno = 0;
   const char *__pyx_filename = NULL;
   int __pyx_clineno = 0;
@@ -6074,21 +7420,21 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
   __pyx_pybuffernd_X_argsorted.rcbuffer = &__pyx_pybuffer_X_argsorted;
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X.rcbuffer->pybuffer, (PyObject*)__pyx_v_X, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_F_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X.rcbuffer->pybuffer, (PyObject*)__pyx_v_X, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_F_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_X.diminfo[0].strides = __pyx_pybuffernd_X.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_X.diminfo[0].shape = __pyx_pybuffernd_X.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_X.diminfo[1].strides = __pyx_pybuffernd_X.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_X.diminfo[1].shape = __pyx_pybuffernd_X.rcbuffer->pybuffer.shape[1];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_y.rcbuffer->pybuffer, (PyObject*)__pyx_v_y, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_y.rcbuffer->pybuffer, (PyObject*)__pyx_v_y, &__Pyx_TypeInfo_nn___pyx_t_7sklearn_4tree_5_tree_DTYPE_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
-  __pyx_pybuffernd_y.diminfo[0].strides = __pyx_pybuffernd_y.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_y.diminfo[0].shape = __pyx_pybuffernd_y.rcbuffer->pybuffer.shape[0];
+  __pyx_pybuffernd_y.diminfo[0].strides = __pyx_pybuffernd_y.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_y.diminfo[0].shape = __pyx_pybuffernd_y.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_y.diminfo[1].strides = __pyx_pybuffernd_y.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_y.diminfo[1].shape = __pyx_pybuffernd_y.rcbuffer->pybuffer.shape[1];
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer, (PyObject*)__pyx_v_X_argsorted, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_F_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer, (PyObject*)__pyx_v_X_argsorted, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_F_CONTIGUOUS, 2, 0, __pyx_stack) == -1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
   __pyx_pybuffernd_X_argsorted.diminfo[0].strides = __pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_X_argsorted.diminfo[0].shape = __pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer.shape[0]; __pyx_pybuffernd_X_argsorted.diminfo[1].strides = __pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer.strides[1]; __pyx_pybuffernd_X_argsorted.diminfo[1].shape = __pyx_pybuffernd_X_argsorted.rcbuffer->pybuffer.shape[1];
 
-  /* "sklearn/tree/_tree.pyx":737
+  /* "sklearn/tree/_tree.pyx":941
  *     """
  *     # Variables
  *     cdef int n_total_samples = X.shape[0]             # <<<<<<<<<<<<<<
@@ -6097,7 +7443,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
   __pyx_v_n_total_samples = (__pyx_v_X->dimensions[0]);
 
-  /* "sklearn/tree/_tree.pyx":738
+  /* "sklearn/tree/_tree.pyx":942
  *     # Variables
  *     cdef int n_total_samples = X.shape[0]
  *     cdef int n_features = X.shape[1]             # <<<<<<<<<<<<<<
@@ -6106,7 +7452,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
   __pyx_v_n_features = (__pyx_v_X->dimensions[1]);
 
-  /* "sklearn/tree/_tree.pyx":739
+  /* "sklearn/tree/_tree.pyx":943
  *     cdef int n_total_samples = X.shape[0]
  *     cdef int n_features = X.shape[1]
  *     cdef int i, a, b, c, n_left, best_i = -1             # <<<<<<<<<<<<<<
@@ -6115,106 +7461,101 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
   __pyx_v_best_i = -1;
 
-  /* "sklearn/tree/_tree.pyx":740
+  /* "sklearn/tree/_tree.pyx":944
  *     cdef int n_features = X.shape[1]
  *     cdef int i, a, b, c, n_left, best_i = -1
  *     cdef np.int32_t feature_idx = -1             # <<<<<<<<<<<<<<
  *     cdef DTYPE_t t, initial_error, error
- *     cdef DTYPE_t best_error = np.inf, best_t = np.inf
+ *     cdef DTYPE_t best_error = INFINITY, best_t = INFINITY
  */
   __pyx_v_feature_idx = -1;
 
-  /* "sklearn/tree/_tree.pyx":742
+  /* "sklearn/tree/_tree.pyx":946
  *     cdef np.int32_t feature_idx = -1
  *     cdef DTYPE_t t, initial_error, error
- *     cdef DTYPE_t best_error = np.inf, best_t = np.inf             # <<<<<<<<<<<<<<
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
- *     cdef DTYPE_t *X_i = NULL
+ *     cdef DTYPE_t best_error = INFINITY, best_t = INFINITY             # <<<<<<<<<<<<<<
+ *     cdef DTYPE_t* X_i = NULL
+ *     cdef int* X_argsorted_i = NULL
  */
-  __pyx_t_1 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 742; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_2 = PyObject_GetAttr(__pyx_t_1, __pyx_n_s__inf); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 742; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __pyx_t_3 = __pyx_PyFloat_AsFloat(__pyx_t_2); if (unlikely((__pyx_t_3 == (npy_float32)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 742; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  __pyx_v_best_error = __pyx_t_3;
-  __pyx_t_2 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 742; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __pyx_t_1 = PyObject_GetAttr(__pyx_t_2, __pyx_n_s__inf); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 742; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  __pyx_t_3 = __pyx_PyFloat_AsFloat(__pyx_t_1); if (unlikely((__pyx_t_3 == (npy_float32)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 742; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __pyx_v_best_t = __pyx_t_3;
+  __pyx_v_best_error = __pyx_v_7sklearn_4tree_5_tree_INFINITY;
+  __pyx_v_best_t = __pyx_v_7sklearn_4tree_5_tree_INFINITY;
 
-  /* "sklearn/tree/_tree.pyx":743
+  /* "sklearn/tree/_tree.pyx":947
  *     cdef DTYPE_t t, initial_error, error
- *     cdef DTYPE_t best_error = np.inf, best_t = np.inf
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data             # <<<<<<<<<<<<<<
- *     cdef DTYPE_t *X_i = NULL
- *     cdef int *X_argsorted_i = NULL
- */
-  __pyx_v_y_ptr = ((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *)__pyx_v_y->data);
-
-  /* "sklearn/tree/_tree.pyx":744
- *     cdef DTYPE_t best_error = np.inf, best_t = np.inf
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
- *     cdef DTYPE_t *X_i = NULL             # <<<<<<<<<<<<<<
- *     cdef int *X_argsorted_i = NULL
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
+ *     cdef DTYPE_t best_error = INFINITY, best_t = INFINITY
+ *     cdef DTYPE_t* X_i = NULL             # <<<<<<<<<<<<<<
+ *     cdef int* X_argsorted_i = NULL
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
  */
   __pyx_v_X_i = NULL;
 
-  /* "sklearn/tree/_tree.pyx":745
- *     cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
- *     cdef DTYPE_t *X_i = NULL
- *     cdef int *X_argsorted_i = NULL             # <<<<<<<<<<<<<<
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
- *     cdef np.ndarray[np.int32_t, ndim=1, mode='c'] features = None
+  /* "sklearn/tree/_tree.pyx":948
+ *     cdef DTYPE_t best_error = INFINITY, best_t = INFINITY
+ *     cdef DTYPE_t* X_i = NULL
+ *     cdef int* X_argsorted_i = NULL             # <<<<<<<<<<<<<<
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+ *     cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data
  */
   __pyx_v_X_argsorted_i = NULL;
 
-  /* "sklearn/tree/_tree.pyx":746
- *     cdef DTYPE_t *X_i = NULL
- *     cdef int *X_argsorted_i = NULL
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data             # <<<<<<<<<<<<<<
- *     cdef np.ndarray[np.int32_t, ndim=1, mode='c'] features = None
+  /* "sklearn/tree/_tree.pyx":949
+ *     cdef DTYPE_t* X_i = NULL
+ *     cdef int* X_argsorted_i = NULL
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data             # <<<<<<<<<<<<<<
+ *     cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data
+ *     cdef np.ndarray[np.int32_t, ndim=1, mode="c"] features = None
+ */
+  __pyx_v_y_ptr = ((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *)__pyx_v_y->data);
+
+  /* "sklearn/tree/_tree.pyx":950
+ *     cdef int* X_argsorted_i = NULL
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+ *     cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data             # <<<<<<<<<<<<<<
+ *     cdef np.ndarray[np.int32_t, ndim=1, mode="c"] features = None
  * 
  */
   __pyx_v_sample_mask_ptr = ((__pyx_t_7sklearn_4tree_5_tree_BOOL_t *)__pyx_v_sample_mask->data);
 
-  /* "sklearn/tree/_tree.pyx":747
- *     cdef int *X_argsorted_i = NULL
- *     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
- *     cdef np.ndarray[np.int32_t, ndim=1, mode='c'] features = None             # <<<<<<<<<<<<<<
+  /* "sklearn/tree/_tree.pyx":951
+ *     cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+ *     cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data
+ *     cdef np.ndarray[np.int32_t, ndim=1, mode="c"] features = None             # <<<<<<<<<<<<<<
  * 
  *     # Compute the column strides (increment in pointer elements to get
  */
-  __pyx_t_4 = ((PyArrayObject *)Py_None);
+  __pyx_t_1 = ((PyArrayObject *)Py_None);
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
-    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_4, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {
+    if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_1, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {
       __pyx_v_features = ((PyArrayObject *)Py_None); __Pyx_INCREF(Py_None); __pyx_pybuffernd_features.rcbuffer->pybuffer.buf = NULL;
-      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 747; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      {__pyx_filename = __pyx_f[0]; __pyx_lineno = 951; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     } else {__pyx_pybuffernd_features.diminfo[0].strides = __pyx_pybuffernd_features.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_features.diminfo[0].shape = __pyx_pybuffernd_features.rcbuffer->pybuffer.shape[0];
     }
   }
-  __pyx_t_4 = 0;
+  __pyx_t_1 = 0;
   __Pyx_INCREF(Py_None);
   __pyx_v_features = ((PyArrayObject *)Py_None);
 
-  /* "sklearn/tree/_tree.pyx":751
+  /* "sklearn/tree/_tree.pyx":955
  *     # Compute the column strides (increment in pointer elements to get
  *     # from column i to i + 1) for `X` and `X_argsorted`
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]             # <<<<<<<<<<<<<<
+ *     cdef int X_elem_stride = X.strides[0]
+ *     cdef int X_col_stride = X.strides[1]
+ */
+  __pyx_v_y_stride = (((int)(__pyx_v_y->strides[0])) / ((int)(__pyx_v_y->strides[1])));
+
+  /* "sklearn/tree/_tree.pyx":956
+ *     # from column i to i + 1) for `X` and `X_argsorted`
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
  *     cdef int X_elem_stride = X.strides[0]             # <<<<<<<<<<<<<<
  *     cdef int X_col_stride = X.strides[1]
  *     cdef int X_stride = X_col_stride / X_elem_stride
  */
   __pyx_v_X_elem_stride = (__pyx_v_X->strides[0]);
 
-  /* "sklearn/tree/_tree.pyx":752
- *     # from column i to i + 1) for `X` and `X_argsorted`
+  /* "sklearn/tree/_tree.pyx":957
+ *     cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
  *     cdef int X_elem_stride = X.strides[0]
  *     cdef int X_col_stride = X.strides[1]             # <<<<<<<<<<<<<<
  *     cdef int X_stride = X_col_stride / X_elem_stride
@@ -6222,7 +7563,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
   __pyx_v_X_col_stride = (__pyx_v_X->strides[1]);
 
-  /* "sklearn/tree/_tree.pyx":753
+  /* "sklearn/tree/_tree.pyx":958
  *     cdef int X_elem_stride = X.strides[0]
  *     cdef int X_col_stride = X.strides[1]
  *     cdef int X_stride = X_col_stride / X_elem_stride             # <<<<<<<<<<<<<<
@@ -6231,7 +7572,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
   __pyx_v_X_stride = (__pyx_v_X_col_stride / __pyx_v_X_elem_stride);
 
-  /* "sklearn/tree/_tree.pyx":754
+  /* "sklearn/tree/_tree.pyx":959
  *     cdef int X_col_stride = X.strides[1]
  *     cdef int X_stride = X_col_stride / X_elem_stride
  *     cdef int X_argsorted_elem_stride = X_argsorted.strides[0]             # <<<<<<<<<<<<<<
@@ -6240,7 +7581,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
   __pyx_v_X_argsorted_elem_stride = (__pyx_v_X_argsorted->strides[0]);
 
-  /* "sklearn/tree/_tree.pyx":755
+  /* "sklearn/tree/_tree.pyx":960
  *     cdef int X_stride = X_col_stride / X_elem_stride
  *     cdef int X_argsorted_elem_stride = X_argsorted.strides[0]
  *     cdef int X_argsorted_col_stride = X_argsorted.strides[1]             # <<<<<<<<<<<<<<
@@ -6249,7 +7590,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
   __pyx_v_X_argsorted_col_stride = (__pyx_v_X_argsorted->strides[1]);
 
-  /* "sklearn/tree/_tree.pyx":756
+  /* "sklearn/tree/_tree.pyx":961
  *     cdef int X_argsorted_elem_stride = X_argsorted.strides[0]
  *     cdef int X_argsorted_col_stride = X_argsorted.strides[1]
  *     cdef int X_argsorted_stride = X_argsorted_col_stride / X_argsorted_elem_stride             # <<<<<<<<<<<<<<
@@ -6258,44 +7599,44 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
   __pyx_v_X_argsorted_stride = (__pyx_v_X_argsorted_col_stride / __pyx_v_X_argsorted_elem_stride);
 
-  /* "sklearn/tree/_tree.pyx":759
+  /* "sklearn/tree/_tree.pyx":964
  * 
  *     # Compute the initial criterion value
- *     X_argsorted_i = <int *>X_argsorted.data             # <<<<<<<<<<<<<<
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
+ *     X_argsorted_i = <int*>X_argsorted.data             # <<<<<<<<<<<<<<
+ *     criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)
  *     initial_error = criterion.eval()
  */
   __pyx_v_X_argsorted_i = ((int *)__pyx_v_X_argsorted->data);
 
-  /* "sklearn/tree/_tree.pyx":760
+  /* "sklearn/tree/_tree.pyx":965
  *     # Compute the initial criterion value
- *     X_argsorted_i = <int *>X_argsorted.data
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)             # <<<<<<<<<<<<<<
+ *     X_argsorted_i = <int*>X_argsorted.data
+ *     criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)             # <<<<<<<<<<<<<<
  *     initial_error = criterion.eval()
  * 
  */
-  ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->init(__pyx_v_criterion, __pyx_v_y_ptr, __pyx_v_sample_mask_ptr, __pyx_v_n_samples, __pyx_v_n_total_samples);
+  ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->init(__pyx_v_criterion, __pyx_v_y_ptr, __pyx_v_y_stride, __pyx_v_sample_mask_ptr, __pyx_v_n_samples, __pyx_v_n_total_samples);
 
-  /* "sklearn/tree/_tree.pyx":761
- *     X_argsorted_i = <int *>X_argsorted.data
- *     criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
+  /* "sklearn/tree/_tree.pyx":966
+ *     X_argsorted_i = <int*>X_argsorted.data
+ *     criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)
  *     initial_error = criterion.eval()             # <<<<<<<<<<<<<<
  * 
  *     if initial_error == 0:  # break early if the node is pure
  */
   __pyx_v_initial_error = ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->eval(__pyx_v_criterion);
 
-  /* "sklearn/tree/_tree.pyx":763
+  /* "sklearn/tree/_tree.pyx":968
  *     initial_error = criterion.eval()
  * 
  *     if initial_error == 0:  # break early if the node is pure             # <<<<<<<<<<<<<<
  *         return best_i, best_t, best_error, initial_error
  * 
  */
-  __pyx_t_5 = (__pyx_v_initial_error == 0.0);
-  if (__pyx_t_5) {
+  __pyx_t_2 = (__pyx_v_initial_error == 0.0);
+  if (__pyx_t_2) {
 
-    /* "sklearn/tree/_tree.pyx":764
+    /* "sklearn/tree/_tree.pyx":969
  * 
  *     if initial_error == 0:  # break early if the node is pure
  *         return best_i, best_t, best_error, initial_error             # <<<<<<<<<<<<<<
@@ -6303,36 +7644,36 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  *     best_error = initial_error
  */
     __Pyx_XDECREF(__pyx_r);
-    __pyx_t_1 = PyInt_FromLong(__pyx_v_best_i); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 764; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_1);
-    __pyx_t_2 = PyFloat_FromDouble(__pyx_v_best_t); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 764; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_2);
-    __pyx_t_6 = PyFloat_FromDouble(__pyx_v_best_error); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 764; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_3 = PyInt_FromLong(__pyx_v_best_i); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 969; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_3);
+    __pyx_t_4 = PyFloat_FromDouble(__pyx_v_best_t); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 969; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_4);
+    __pyx_t_5 = PyFloat_FromDouble(__pyx_v_best_error); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 969; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_5);
+    __pyx_t_6 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 969; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     __Pyx_GOTREF(__pyx_t_6);
-    __pyx_t_7 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 764; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_7 = PyTuple_New(4); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 969; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     __Pyx_GOTREF(__pyx_t_7);
-    __pyx_t_8 = PyTuple_New(4); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 764; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_8);
-    PyTuple_SET_ITEM(__pyx_t_8, 0, __pyx_t_1);
-    __Pyx_GIVEREF(__pyx_t_1);
-    PyTuple_SET_ITEM(__pyx_t_8, 1, __pyx_t_2);
-    __Pyx_GIVEREF(__pyx_t_2);
-    PyTuple_SET_ITEM(__pyx_t_8, 2, __pyx_t_6);
+    PyTuple_SET_ITEM(__pyx_t_7, 0, __pyx_t_3);
+    __Pyx_GIVEREF(__pyx_t_3);
+    PyTuple_SET_ITEM(__pyx_t_7, 1, __pyx_t_4);
+    __Pyx_GIVEREF(__pyx_t_4);
+    PyTuple_SET_ITEM(__pyx_t_7, 2, __pyx_t_5);
+    __Pyx_GIVEREF(__pyx_t_5);
+    PyTuple_SET_ITEM(__pyx_t_7, 3, __pyx_t_6);
     __Pyx_GIVEREF(__pyx_t_6);
-    PyTuple_SET_ITEM(__pyx_t_8, 3, __pyx_t_7);
-    __Pyx_GIVEREF(__pyx_t_7);
-    __pyx_t_1 = 0;
-    __pyx_t_2 = 0;
+    __pyx_t_3 = 0;
+    __pyx_t_4 = 0;
+    __pyx_t_5 = 0;
     __pyx_t_6 = 0;
+    __pyx_r = ((PyObject *)__pyx_t_7);
     __pyx_t_7 = 0;
-    __pyx_r = ((PyObject *)__pyx_t_8);
-    __pyx_t_8 = 0;
     goto __pyx_L0;
     goto __pyx_L3;
   }
   __pyx_L3:;
 
-  /* "sklearn/tree/_tree.pyx":766
+  /* "sklearn/tree/_tree.pyx":971
  *         return best_i, best_t, best_error, initial_error
  * 
  *     best_error = initial_error             # <<<<<<<<<<<<<<
@@ -6341,79 +7682,79 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
   __pyx_v_best_error = __pyx_v_initial_error;
 
-  /* "sklearn/tree/_tree.pyx":769
+  /* "sklearn/tree/_tree.pyx":974
  * 
  *     # Features to consider
  *     features = np.arange(n_features, dtype=np.int32)             # <<<<<<<<<<<<<<
  *     if max_features < 0 or max_features >= n_features:
  *         max_features = n_features
  */
-  __pyx_t_8 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_8);
-  __pyx_t_7 = PyObject_GetAttr(__pyx_t_8, __pyx_n_s__arange); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_7 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_7);
-  __Pyx_DECREF(__pyx_t_8); __pyx_t_8 = 0;
-  __pyx_t_8 = PyInt_FromLong(__pyx_v_n_features); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_8);
-  __pyx_t_6 = PyTuple_New(1); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_6 = PyObject_GetAttr(__pyx_t_7, __pyx_n_s__arange); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_6);
-  PyTuple_SET_ITEM(__pyx_t_6, 0, __pyx_t_8);
-  __Pyx_GIVEREF(__pyx_t_8);
-  __pyx_t_8 = 0;
-  __pyx_t_8 = PyDict_New(); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(((PyObject *)__pyx_t_8));
-  __pyx_t_2 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  __pyx_t_1 = PyObject_GetAttr(__pyx_t_2, __pyx_n_s__int32); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
-  if (PyDict_SetItem(__pyx_t_8, ((PyObject *)__pyx_n_s__dtype), __pyx_t_1) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-  __pyx_t_1 = PyObject_Call(__pyx_t_7, ((PyObject *)__pyx_t_6), ((PyObject *)__pyx_t_8)); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
   __Pyx_DECREF(__pyx_t_7); __pyx_t_7 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_6)); __pyx_t_6 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_t_8)); __pyx_t_8 = 0;
-  if (!(likely(((__pyx_t_1) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_1, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __pyx_t_4 = ((PyArrayObject *)__pyx_t_1);
+  __pyx_t_7 = PyInt_FromLong(__pyx_v_n_features); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_7);
+  __pyx_t_5 = PyTuple_New(1); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_5);
+  PyTuple_SET_ITEM(__pyx_t_5, 0, __pyx_t_7);
+  __Pyx_GIVEREF(__pyx_t_7);
+  __pyx_t_7 = 0;
+  __pyx_t_7 = PyDict_New(); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(((PyObject *)__pyx_t_7));
+  __pyx_t_4 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_4);
+  __pyx_t_3 = PyObject_GetAttr(__pyx_t_4, __pyx_n_s__int32); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_3);
+  __Pyx_DECREF(__pyx_t_4); __pyx_t_4 = 0;
+  if (PyDict_SetItem(__pyx_t_7, ((PyObject *)__pyx_n_s__dtype), __pyx_t_3) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_3); __pyx_t_3 = 0;
+  __pyx_t_3 = PyObject_Call(__pyx_t_6, ((PyObject *)__pyx_t_5), ((PyObject *)__pyx_t_7)); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_3);
+  __Pyx_DECREF(__pyx_t_6); __pyx_t_6 = 0;
+  __Pyx_DECREF(((PyObject *)__pyx_t_5)); __pyx_t_5 = 0;
+  __Pyx_DECREF(((PyObject *)__pyx_t_7)); __pyx_t_7 = 0;
+  if (!(likely(((__pyx_t_3) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_3, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = ((PyArrayObject *)__pyx_t_3);
   {
     __Pyx_BufFmt_StackElem __pyx_stack[1];
     __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_features.rcbuffer->pybuffer);
-    __pyx_t_9 = __Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_4, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack);
-    if (unlikely(__pyx_t_9 < 0)) {
-      PyErr_Fetch(&__pyx_t_10, &__pyx_t_11, &__pyx_t_12);
+    __pyx_t_8 = __Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_1, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack);
+    if (unlikely(__pyx_t_8 < 0)) {
+      PyErr_Fetch(&__pyx_t_9, &__pyx_t_10, &__pyx_t_11);
       if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_v_features, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {
-        Py_XDECREF(__pyx_t_10); Py_XDECREF(__pyx_t_11); Py_XDECREF(__pyx_t_12);
+        Py_XDECREF(__pyx_t_9); Py_XDECREF(__pyx_t_10); Py_XDECREF(__pyx_t_11);
         __Pyx_RaiseBufferFallbackError();
       } else {
-        PyErr_Restore(__pyx_t_10, __pyx_t_11, __pyx_t_12);
+        PyErr_Restore(__pyx_t_9, __pyx_t_10, __pyx_t_11);
       }
     }
     __pyx_pybuffernd_features.diminfo[0].strides = __pyx_pybuffernd_features.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_features.diminfo[0].shape = __pyx_pybuffernd_features.rcbuffer->pybuffer.shape[0];
-    if (unlikely(__pyx_t_9 < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 769; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    if (unlikely(__pyx_t_8 < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 974; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   }
-  __pyx_t_4 = 0;
-  __Pyx_DECREF(((PyObject *)__pyx_v_features));
-  __pyx_v_features = ((PyArrayObject *)__pyx_t_1);
   __pyx_t_1 = 0;
+  __Pyx_DECREF(((PyObject *)__pyx_v_features));
+  __pyx_v_features = ((PyArrayObject *)__pyx_t_3);
+  __pyx_t_3 = 0;
 
-  /* "sklearn/tree/_tree.pyx":770
+  /* "sklearn/tree/_tree.pyx":975
  *     # Features to consider
  *     features = np.arange(n_features, dtype=np.int32)
  *     if max_features < 0 or max_features >= n_features:             # <<<<<<<<<<<<<<
  *         max_features = n_features
  *     else:
  */
-  __pyx_t_5 = (__pyx_v_max_features < 0);
-  if (!__pyx_t_5) {
-    __pyx_t_13 = (__pyx_v_max_features >= __pyx_v_n_features);
-    __pyx_t_14 = __pyx_t_13;
+  __pyx_t_2 = (__pyx_v_max_features < 0);
+  if (!__pyx_t_2) {
+    __pyx_t_12 = (__pyx_v_max_features >= __pyx_v_n_features);
+    __pyx_t_13 = __pyx_t_12;
   } else {
-    __pyx_t_14 = __pyx_t_5;
+    __pyx_t_13 = __pyx_t_2;
   }
-  if (__pyx_t_14) {
+  if (__pyx_t_13) {
 
-    /* "sklearn/tree/_tree.pyx":771
+    /* "sklearn/tree/_tree.pyx":976
  *     features = np.arange(n_features, dtype=np.int32)
  *     if max_features < 0 or max_features >= n_features:
  *         max_features = n_features             # <<<<<<<<<<<<<<
@@ -6425,92 +7766,91 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
   }
   /*else*/ {
 
-    /* "sklearn/tree/_tree.pyx":773
+    /* "sklearn/tree/_tree.pyx":978
  *         max_features = n_features
  *     else:
  *         features = random_state.permutation(features)[:max_features]             # <<<<<<<<<<<<<<
  * 
  *     # Look for the best random split
  */
-    __pyx_t_1 = PyObject_GetAttr(__pyx_v_random_state, __pyx_n_s__permutation); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 773; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_1);
-    __pyx_t_8 = PyTuple_New(1); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 773; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_8);
+    __pyx_t_3 = PyObject_GetAttr(__pyx_v_random_state, __pyx_n_s__permutation); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 978; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_3);
+    __pyx_t_7 = PyTuple_New(1); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 978; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_7);
     __Pyx_INCREF(((PyObject *)__pyx_v_features));
-    PyTuple_SET_ITEM(__pyx_t_8, 0, ((PyObject *)__pyx_v_features));
+    PyTuple_SET_ITEM(__pyx_t_7, 0, ((PyObject *)__pyx_v_features));
     __Pyx_GIVEREF(((PyObject *)__pyx_v_features));
-    __pyx_t_6 = PyObject_Call(__pyx_t_1, ((PyObject *)__pyx_t_8), NULL); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 773; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_6);
-    __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-    __Pyx_DECREF(((PyObject *)__pyx_t_8)); __pyx_t_8 = 0;
-    __pyx_t_8 = __Pyx_PySequence_GetSlice(__pyx_t_6, 0, __pyx_v_max_features); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 773; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_8);
-    __Pyx_DECREF(__pyx_t_6); __pyx_t_6 = 0;
-    if (!(likely(((__pyx_t_8) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_8, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 773; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __pyx_t_4 = ((PyArrayObject *)__pyx_t_8);
+    __pyx_t_5 = PyObject_Call(__pyx_t_3, ((PyObject *)__pyx_t_7), NULL); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 978; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_5);
+    __Pyx_DECREF(__pyx_t_3); __pyx_t_3 = 0;
+    __Pyx_DECREF(((PyObject *)__pyx_t_7)); __pyx_t_7 = 0;
+    __pyx_t_7 = __Pyx_PySequence_GetSlice(__pyx_t_5, 0, __pyx_v_max_features); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 978; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_7);
+    __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
+    if (!(likely(((__pyx_t_7) == Py_None) || likely(__Pyx_TypeTest(__pyx_t_7, __pyx_ptype_5numpy_ndarray))))) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 978; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_1 = ((PyArrayObject *)__pyx_t_7);
     {
       __Pyx_BufFmt_StackElem __pyx_stack[1];
       __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_features.rcbuffer->pybuffer);
-      __pyx_t_9 = __Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_4, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack);
-      if (unlikely(__pyx_t_9 < 0)) {
-        PyErr_Fetch(&__pyx_t_12, &__pyx_t_11, &__pyx_t_10);
+      __pyx_t_8 = __Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_t_1, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack);
+      if (unlikely(__pyx_t_8 < 0)) {
+        PyErr_Fetch(&__pyx_t_11, &__pyx_t_10, &__pyx_t_9);
         if (unlikely(__Pyx_GetBufferAndValidate(&__pyx_pybuffernd_features.rcbuffer->pybuffer, (PyObject*)__pyx_v_features, &__Pyx_TypeInfo_nn___pyx_t_5numpy_int32_t, PyBUF_FORMAT| PyBUF_C_CONTIGUOUS, 1, 0, __pyx_stack) == -1)) {
-          Py_XDECREF(__pyx_t_12); Py_XDECREF(__pyx_t_11); Py_XDECREF(__pyx_t_10);
+          Py_XDECREF(__pyx_t_11); Py_XDECREF(__pyx_t_10); Py_XDECREF(__pyx_t_9);
           __Pyx_RaiseBufferFallbackError();
         } else {
-          PyErr_Restore(__pyx_t_12, __pyx_t_11, __pyx_t_10);
+          PyErr_Restore(__pyx_t_11, __pyx_t_10, __pyx_t_9);
         }
       }
       __pyx_pybuffernd_features.diminfo[0].strides = __pyx_pybuffernd_features.rcbuffer->pybuffer.strides[0]; __pyx_pybuffernd_features.diminfo[0].shape = __pyx_pybuffernd_features.rcbuffer->pybuffer.shape[0];
-      if (unlikely(__pyx_t_9 < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 773; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+      if (unlikely(__pyx_t_8 < 0)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 978; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     }
-    __pyx_t_4 = 0;
+    __pyx_t_1 = 0;
     __Pyx_DECREF(((PyObject *)__pyx_v_features));
-    __pyx_v_features = ((PyArrayObject *)__pyx_t_8);
-    __pyx_t_8 = 0;
+    __pyx_v_features = ((PyArrayObject *)__pyx_t_7);
+    __pyx_t_7 = 0;
   }
   __pyx_L4:;
 
-  /* "sklearn/tree/_tree.pyx":776
+  /* "sklearn/tree/_tree.pyx":981
  * 
  *     # Look for the best random split
- *     for feature_idx in range(max_features):             # <<<<<<<<<<<<<<
+ *     for feature_idx from 0 <= feature_idx < max_features:             # <<<<<<<<<<<<<<
  *         i = features[feature_idx]
  *         # Get i-th col of X and X_sorted
  */
-  __pyx_t_9 = __pyx_v_max_features;
-  for (__pyx_t_15 = 0; __pyx_t_15 < __pyx_t_9; __pyx_t_15+=1) {
-    __pyx_v_feature_idx = __pyx_t_15;
+  __pyx_t_8 = __pyx_v_max_features;
+  for (__pyx_v_feature_idx = 0; __pyx_v_feature_idx < __pyx_t_8; __pyx_v_feature_idx++) {
 
-    /* "sklearn/tree/_tree.pyx":777
+    /* "sklearn/tree/_tree.pyx":982
  *     # Look for the best random split
- *     for feature_idx in range(max_features):
+ *     for feature_idx from 0 <= feature_idx < max_features:
  *         i = features[feature_idx]             # <<<<<<<<<<<<<<
  *         # Get i-th col of X and X_sorted
- *         X_i = (<DTYPE_t *>X.data) + X_stride * i
+ *         X_i = (<DTYPE_t*>X.data) + X_stride * i
  */
-    __pyx_t_16 = __pyx_v_feature_idx;
-    __pyx_v_i = (*__Pyx_BufPtrCContig1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_features.rcbuffer->pybuffer.buf, __pyx_t_16, __pyx_pybuffernd_features.diminfo[0].strides));
+    __pyx_t_14 = __pyx_v_feature_idx;
+    __pyx_v_i = (*__Pyx_BufPtrCContig1d(__pyx_t_5numpy_int32_t *, __pyx_pybuffernd_features.rcbuffer->pybuffer.buf, __pyx_t_14, __pyx_pybuffernd_features.diminfo[0].strides));
 
-    /* "sklearn/tree/_tree.pyx":779
+    /* "sklearn/tree/_tree.pyx":984
  *         i = features[feature_idx]
  *         # Get i-th col of X and X_sorted
- *         X_i = (<DTYPE_t *>X.data) + X_stride * i             # <<<<<<<<<<<<<<
- *         X_argsorted_i = (<int *>X_argsorted.data) + X_argsorted_stride * i
+ *         X_i = (<DTYPE_t*>X.data) + X_stride * i             # <<<<<<<<<<<<<<
+ *         X_argsorted_i = (<int*>X_argsorted.data) + X_argsorted_stride * i
  * 
  */
     __pyx_v_X_i = (((__pyx_t_7sklearn_4tree_5_tree_DTYPE_t *)__pyx_v_X->data) + (__pyx_v_X_stride * __pyx_v_i));
 
-    /* "sklearn/tree/_tree.pyx":780
+    /* "sklearn/tree/_tree.pyx":985
  *         # Get i-th col of X and X_sorted
- *         X_i = (<DTYPE_t *>X.data) + X_stride * i
- *         X_argsorted_i = (<int *>X_argsorted.data) + X_argsorted_stride * i             # <<<<<<<<<<<<<<
+ *         X_i = (<DTYPE_t*>X.data) + X_stride * i
+ *         X_argsorted_i = (<int*>X_argsorted.data) + X_argsorted_stride * i             # <<<<<<<<<<<<<<
  * 
  *         # Reset the criterion for this feature
  */
     __pyx_v_X_argsorted_i = (((int *)__pyx_v_X_argsorted->data) + (__pyx_v_X_argsorted_stride * __pyx_v_i));
 
-    /* "sklearn/tree/_tree.pyx":783
+    /* "sklearn/tree/_tree.pyx":988
  * 
  *         # Reset the criterion for this feature
  *         criterion.reset()             # <<<<<<<<<<<<<<
@@ -6519,7 +7859,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
     ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->reset(__pyx_v_criterion);
 
-    /* "sklearn/tree/_tree.pyx":786
+    /* "sklearn/tree/_tree.pyx":991
  * 
  *         # Find min and max
  *         a = 0             # <<<<<<<<<<<<<<
@@ -6528,7 +7868,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
     __pyx_v_a = 0;
 
-    /* "sklearn/tree/_tree.pyx":787
+    /* "sklearn/tree/_tree.pyx":992
  *         # Find min and max
  *         a = 0
  *         while sample_mask_ptr[X_argsorted_i[a]] == 0:             # <<<<<<<<<<<<<<
@@ -6536,10 +7876,10 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  * 
  */
     while (1) {
-      __pyx_t_14 = ((__pyx_v_sample_mask_ptr[(__pyx_v_X_argsorted_i[__pyx_v_a])]) == 0);
-      if (!__pyx_t_14) break;
+      __pyx_t_13 = ((__pyx_v_sample_mask_ptr[(__pyx_v_X_argsorted_i[__pyx_v_a])]) == 0);
+      if (!__pyx_t_13) break;
 
-      /* "sklearn/tree/_tree.pyx":788
+      /* "sklearn/tree/_tree.pyx":993
  *         a = 0
  *         while sample_mask_ptr[X_argsorted_i[a]] == 0:
  *             a = a + 1             # <<<<<<<<<<<<<<
@@ -6549,7 +7889,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
       __pyx_v_a = (__pyx_v_a + 1);
     }
 
-    /* "sklearn/tree/_tree.pyx":790
+    /* "sklearn/tree/_tree.pyx":995
  *             a = a + 1
  * 
  *         b = n_total_samples - 1             # <<<<<<<<<<<<<<
@@ -6558,7 +7898,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
     __pyx_v_b = (__pyx_v_n_total_samples - 1);
 
-    /* "sklearn/tree/_tree.pyx":791
+    /* "sklearn/tree/_tree.pyx":996
  * 
  *         b = n_total_samples - 1
  *         while sample_mask_ptr[X_argsorted_i[b]] == 0:             # <<<<<<<<<<<<<<
@@ -6566,10 +7906,10 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  * 
  */
     while (1) {
-      __pyx_t_14 = ((__pyx_v_sample_mask_ptr[(__pyx_v_X_argsorted_i[__pyx_v_b])]) == 0);
-      if (!__pyx_t_14) break;
+      __pyx_t_13 = ((__pyx_v_sample_mask_ptr[(__pyx_v_X_argsorted_i[__pyx_v_b])]) == 0);
+      if (!__pyx_t_13) break;
 
-      /* "sklearn/tree/_tree.pyx":792
+      /* "sklearn/tree/_tree.pyx":997
  *         b = n_total_samples - 1
  *         while sample_mask_ptr[X_argsorted_i[b]] == 0:
  *             b = b - 1             # <<<<<<<<<<<<<<
@@ -6579,23 +7919,23 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
       __pyx_v_b = (__pyx_v_b - 1);
     }
 
-    /* "sklearn/tree/_tree.pyx":794
+    /* "sklearn/tree/_tree.pyx":999
  *             b = b - 1
  * 
  *         if b <= a or X_i[X_argsorted_i[a]] == X_i[X_argsorted_i[b]]:             # <<<<<<<<<<<<<<
  *             continue
  * 
  */
-    __pyx_t_14 = (__pyx_v_b <= __pyx_v_a);
-    if (!__pyx_t_14) {
-      __pyx_t_5 = ((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_a])]) == (__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_b])]));
-      __pyx_t_13 = __pyx_t_5;
+    __pyx_t_13 = (__pyx_v_b <= __pyx_v_a);
+    if (!__pyx_t_13) {
+      __pyx_t_2 = ((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_a])]) == (__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_b])]));
+      __pyx_t_12 = __pyx_t_2;
     } else {
-      __pyx_t_13 = __pyx_t_14;
+      __pyx_t_12 = __pyx_t_13;
     }
-    if (__pyx_t_13) {
+    if (__pyx_t_12) {
 
-      /* "sklearn/tree/_tree.pyx":795
+      /* "sklearn/tree/_tree.pyx":1000
  * 
  *         if b <= a or X_i[X_argsorted_i[a]] == X_i[X_argsorted_i[b]]:
  *             continue             # <<<<<<<<<<<<<<
@@ -6607,53 +7947,53 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
     }
     __pyx_L11:;
 
-    /* "sklearn/tree/_tree.pyx":798
+    /* "sklearn/tree/_tree.pyx":1003
  * 
  *         # Draw a random threshold in [a, b)
  *         t = X_i[X_argsorted_i[a]] + (random_state.rand() *             # <<<<<<<<<<<<<<
  *                                      (X_i[X_argsorted_i[b]] - X_i[X_argsorted_i[a]]))
  *         if t == X_i[X_argsorted_i[b]]:
  */
-    __pyx_t_8 = PyFloat_FromDouble((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_a])])); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 798; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_8);
-    __pyx_t_6 = PyObject_GetAttr(__pyx_v_random_state, __pyx_n_s__rand); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 798; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_6);
-    __pyx_t_1 = PyObject_Call(__pyx_t_6, ((PyObject *)__pyx_empty_tuple), NULL); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 798; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_1);
-    __Pyx_DECREF(__pyx_t_6); __pyx_t_6 = 0;
+    __pyx_t_7 = PyFloat_FromDouble((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_a])])); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1003; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_7);
+    __pyx_t_5 = PyObject_GetAttr(__pyx_v_random_state, __pyx_n_s__rand); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1003; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_5);
+    __pyx_t_3 = PyObject_Call(__pyx_t_5, ((PyObject *)__pyx_empty_tuple), NULL); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1003; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_3);
+    __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
 
-    /* "sklearn/tree/_tree.pyx":799
+    /* "sklearn/tree/_tree.pyx":1004
  *         # Draw a random threshold in [a, b)
  *         t = X_i[X_argsorted_i[a]] + (random_state.rand() *
  *                                      (X_i[X_argsorted_i[b]] - X_i[X_argsorted_i[a]]))             # <<<<<<<<<<<<<<
  *         if t == X_i[X_argsorted_i[b]]:
  *             t = X_i[X_argsorted_i[a]]
  */
-    __pyx_t_6 = PyFloat_FromDouble(((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_b])]) - (__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_a])]))); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 799; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_6);
-    __pyx_t_7 = PyNumber_Multiply(__pyx_t_1, __pyx_t_6); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 798; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-    __Pyx_GOTREF(__pyx_t_7);
-    __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
-    __Pyx_DECREF(__pyx_t_6); __pyx_t_6 = 0;
-    __pyx_t_6 = PyNumber_Add(__pyx_t_8, __pyx_t_7); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 798; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __pyx_t_5 = PyFloat_FromDouble(((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_b])]) - (__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_a])]))); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1004; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_5);
+    __pyx_t_6 = PyNumber_Multiply(__pyx_t_3, __pyx_t_5); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1003; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     __Pyx_GOTREF(__pyx_t_6);
-    __Pyx_DECREF(__pyx_t_8); __pyx_t_8 = 0;
+    __Pyx_DECREF(__pyx_t_3); __pyx_t_3 = 0;
+    __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
+    __pyx_t_5 = PyNumber_Add(__pyx_t_7, __pyx_t_6); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1003; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_GOTREF(__pyx_t_5);
     __Pyx_DECREF(__pyx_t_7); __pyx_t_7 = 0;
-    __pyx_t_3 = __pyx_PyFloat_AsFloat(__pyx_t_6); if (unlikely((__pyx_t_3 == (npy_float32)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 798; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
     __Pyx_DECREF(__pyx_t_6); __pyx_t_6 = 0;
-    __pyx_v_t = __pyx_t_3;
+    __pyx_t_15 = __pyx_PyFloat_AsFloat(__pyx_t_5); if (unlikely((__pyx_t_15 == (npy_float32)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1003; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+    __Pyx_DECREF(__pyx_t_5); __pyx_t_5 = 0;
+    __pyx_v_t = __pyx_t_15;
 
-    /* "sklearn/tree/_tree.pyx":800
+    /* "sklearn/tree/_tree.pyx":1005
  *         t = X_i[X_argsorted_i[a]] + (random_state.rand() *
  *                                      (X_i[X_argsorted_i[b]] - X_i[X_argsorted_i[a]]))
  *         if t == X_i[X_argsorted_i[b]]:             # <<<<<<<<<<<<<<
  *             t = X_i[X_argsorted_i[a]]
  * 
  */
-    __pyx_t_13 = (__pyx_v_t == (__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_b])]));
-    if (__pyx_t_13) {
+    __pyx_t_12 = (__pyx_v_t == (__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_b])]));
+    if (__pyx_t_12) {
 
-      /* "sklearn/tree/_tree.pyx":801
+      /* "sklearn/tree/_tree.pyx":1006
  *                                      (X_i[X_argsorted_i[b]] - X_i[X_argsorted_i[a]]))
  *         if t == X_i[X_argsorted_i[b]]:
  *             t = X_i[X_argsorted_i[a]]             # <<<<<<<<<<<<<<
@@ -6665,7 +8005,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
     }
     __pyx_L12:;
 
-    /* "sklearn/tree/_tree.pyx":804
+    /* "sklearn/tree/_tree.pyx":1009
  * 
  *         # Find the sample just greater than t
  *         c = a + 1             # <<<<<<<<<<<<<<
@@ -6674,7 +8014,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
     __pyx_v_c = (__pyx_v_a + 1);
 
-    /* "sklearn/tree/_tree.pyx":806
+    /* "sklearn/tree/_tree.pyx":1011
  *         c = a + 1
  * 
  *         while True:             # <<<<<<<<<<<<<<
@@ -6684,33 +8024,33 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
     while (1) {
       if (!1) break;
 
-      /* "sklearn/tree/_tree.pyx":807
+      /* "sklearn/tree/_tree.pyx":1012
  * 
  *         while True:
  *             if sample_mask_ptr[X_argsorted_i[c]] != 0:             # <<<<<<<<<<<<<<
  *                 if X_i[X_argsorted_i[c]] > t or c == b:
  *                     break
  */
-      __pyx_t_13 = ((__pyx_v_sample_mask_ptr[(__pyx_v_X_argsorted_i[__pyx_v_c])]) != 0);
-      if (__pyx_t_13) {
+      __pyx_t_12 = ((__pyx_v_sample_mask_ptr[(__pyx_v_X_argsorted_i[__pyx_v_c])]) != 0);
+      if (__pyx_t_12) {
 
-        /* "sklearn/tree/_tree.pyx":808
+        /* "sklearn/tree/_tree.pyx":1013
  *         while True:
  *             if sample_mask_ptr[X_argsorted_i[c]] != 0:
  *                 if X_i[X_argsorted_i[c]] > t or c == b:             # <<<<<<<<<<<<<<
  *                     break
  * 
  */
-        __pyx_t_13 = ((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_c])]) > __pyx_v_t);
-        if (!__pyx_t_13) {
-          __pyx_t_14 = (__pyx_v_c == __pyx_v_b);
-          __pyx_t_5 = __pyx_t_14;
+        __pyx_t_12 = ((__pyx_v_X_i[(__pyx_v_X_argsorted_i[__pyx_v_c])]) > __pyx_v_t);
+        if (!__pyx_t_12) {
+          __pyx_t_13 = (__pyx_v_c == __pyx_v_b);
+          __pyx_t_2 = __pyx_t_13;
         } else {
-          __pyx_t_5 = __pyx_t_13;
+          __pyx_t_2 = __pyx_t_12;
         }
-        if (__pyx_t_5) {
+        if (__pyx_t_2) {
 
-          /* "sklearn/tree/_tree.pyx":809
+          /* "sklearn/tree/_tree.pyx":1014
  *             if sample_mask_ptr[X_argsorted_i[c]] != 0:
  *                 if X_i[X_argsorted_i[c]] > t or c == b:
  *                     break             # <<<<<<<<<<<<<<
@@ -6725,7 +8065,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
       }
       __pyx_L15:;
 
-      /* "sklearn/tree/_tree.pyx":811
+      /* "sklearn/tree/_tree.pyx":1016
  *                     break
  * 
  *             c += 1             # <<<<<<<<<<<<<<
@@ -6736,41 +8076,41 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
     }
     __pyx_L14_break:;
 
-    /* "sklearn/tree/_tree.pyx":814
+    /* "sklearn/tree/_tree.pyx":1019
  * 
  *         # Better than the best so far?
- *         n_left = criterion.update(0, c, y_ptr, X_argsorted_i, sample_mask_ptr)             # <<<<<<<<<<<<<<
+ *         n_left = criterion.update(0, c, y_ptr, y_stride, X_argsorted_i, sample_mask_ptr)             # <<<<<<<<<<<<<<
  *         error = criterion.eval()
  * 
  */
-    __pyx_v_n_left = ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->update(__pyx_v_criterion, 0, __pyx_v_c, __pyx_v_y_ptr, __pyx_v_X_argsorted_i, __pyx_v_sample_mask_ptr);
+    __pyx_v_n_left = ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->update(__pyx_v_criterion, 0, __pyx_v_c, __pyx_v_y_ptr, __pyx_v_y_stride, __pyx_v_X_argsorted_i, __pyx_v_sample_mask_ptr);
 
-    /* "sklearn/tree/_tree.pyx":815
+    /* "sklearn/tree/_tree.pyx":1020
  *         # Better than the best so far?
- *         n_left = criterion.update(0, c, y_ptr, X_argsorted_i, sample_mask_ptr)
+ *         n_left = criterion.update(0, c, y_ptr, y_stride, X_argsorted_i, sample_mask_ptr)
  *         error = criterion.eval()             # <<<<<<<<<<<<<<
  * 
  *         if n_left < min_leaf or (n_samples - n_left) < min_leaf:
  */
     __pyx_v_error = ((struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion *)__pyx_v_criterion->__pyx_vtab)->eval(__pyx_v_criterion);
 
-    /* "sklearn/tree/_tree.pyx":817
+    /* "sklearn/tree/_tree.pyx":1022
  *         error = criterion.eval()
  * 
  *         if n_left < min_leaf or (n_samples - n_left) < min_leaf:             # <<<<<<<<<<<<<<
  *             continue
  * 
  */
-    __pyx_t_5 = (__pyx_v_n_left < __pyx_v_min_leaf);
-    if (!__pyx_t_5) {
-      __pyx_t_13 = ((__pyx_v_n_samples - __pyx_v_n_left) < __pyx_v_min_leaf);
-      __pyx_t_14 = __pyx_t_13;
+    __pyx_t_2 = (__pyx_v_n_left < __pyx_v_min_leaf);
+    if (!__pyx_t_2) {
+      __pyx_t_12 = ((__pyx_v_n_samples - __pyx_v_n_left) < __pyx_v_min_leaf);
+      __pyx_t_13 = __pyx_t_12;
     } else {
-      __pyx_t_14 = __pyx_t_5;
+      __pyx_t_13 = __pyx_t_2;
     }
-    if (__pyx_t_14) {
+    if (__pyx_t_13) {
 
-      /* "sklearn/tree/_tree.pyx":818
+      /* "sklearn/tree/_tree.pyx":1023
  * 
  *         if n_left < min_leaf or (n_samples - n_left) < min_leaf:
  *             continue             # <<<<<<<<<<<<<<
@@ -6782,17 +8122,17 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
     }
     __pyx_L17:;
 
-    /* "sklearn/tree/_tree.pyx":820
+    /* "sklearn/tree/_tree.pyx":1025
  *             continue
  * 
  *         if error < best_error:             # <<<<<<<<<<<<<<
  *             best_i = i
  *             best_t = t
  */
-    __pyx_t_14 = (__pyx_v_error < __pyx_v_best_error);
-    if (__pyx_t_14) {
+    __pyx_t_13 = (__pyx_v_error < __pyx_v_best_error);
+    if (__pyx_t_13) {
 
-      /* "sklearn/tree/_tree.pyx":821
+      /* "sklearn/tree/_tree.pyx":1026
  * 
  *         if error < best_error:
  *             best_i = i             # <<<<<<<<<<<<<<
@@ -6801,7 +8141,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
       __pyx_v_best_i = __pyx_v_i;
 
-      /* "sklearn/tree/_tree.pyx":822
+      /* "sklearn/tree/_tree.pyx":1027
  *         if error < best_error:
  *             best_i = i
  *             best_t = t             # <<<<<<<<<<<<<<
@@ -6810,7 +8150,7 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
  */
       __pyx_v_best_t = __pyx_v_t;
 
-      /* "sklearn/tree/_tree.pyx":823
+      /* "sklearn/tree/_tree.pyx":1028
  *             best_i = i
  *             best_t = t
  *             best_error = error             # <<<<<<<<<<<<<<
@@ -6824,46 +8164,46 @@ static PyObject *__pyx_pf_7sklearn_4tree_5_tree_10_find_best_random_split(CYTHON
     __pyx_L5_continue:;
   }
 
-  /* "sklearn/tree/_tree.pyx":825
+  /* "sklearn/tree/_tree.pyx":1030
  *             best_error = error
  * 
  *     return best_i, best_t, best_error, initial_error             # <<<<<<<<<<<<<<
  */
   __Pyx_XDECREF(__pyx_r);
-  __pyx_t_6 = PyInt_FromLong(__pyx_v_best_i); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 825; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_5 = PyInt_FromLong(__pyx_v_best_i); if (unlikely(!__pyx_t_5)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1030; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_5);
+  __pyx_t_6 = PyFloat_FromDouble(__pyx_v_best_t); if (unlikely(!__pyx_t_6)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1030; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_6);
-  __pyx_t_7 = PyFloat_FromDouble(__pyx_v_best_t); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 825; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_7 = PyFloat_FromDouble(__pyx_v_best_error); if (unlikely(!__pyx_t_7)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1030; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_7);
-  __pyx_t_8 = PyFloat_FromDouble(__pyx_v_best_error); if (unlikely(!__pyx_t_8)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 825; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_8);
-  __pyx_t_1 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 825; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_1);
-  __pyx_t_2 = PyTuple_New(4); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 825; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  PyTuple_SET_ITEM(__pyx_t_2, 0, __pyx_t_6);
+  __pyx_t_3 = PyFloat_FromDouble(__pyx_v_initial_error); if (unlikely(!__pyx_t_3)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1030; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_3);
+  __pyx_t_4 = PyTuple_New(4); if (unlikely(!__pyx_t_4)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1030; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_4);
+  PyTuple_SET_ITEM(__pyx_t_4, 0, __pyx_t_5);
+  __Pyx_GIVEREF(__pyx_t_5);
+  PyTuple_SET_ITEM(__pyx_t_4, 1, __pyx_t_6);
   __Pyx_GIVEREF(__pyx_t_6);
-  PyTuple_SET_ITEM(__pyx_t_2, 1, __pyx_t_7);
+  PyTuple_SET_ITEM(__pyx_t_4, 2, __pyx_t_7);
   __Pyx_GIVEREF(__pyx_t_7);
-  PyTuple_SET_ITEM(__pyx_t_2, 2, __pyx_t_8);
-  __Pyx_GIVEREF(__pyx_t_8);
-  PyTuple_SET_ITEM(__pyx_t_2, 3, __pyx_t_1);
-  __Pyx_GIVEREF(__pyx_t_1);
+  PyTuple_SET_ITEM(__pyx_t_4, 3, __pyx_t_3);
+  __Pyx_GIVEREF(__pyx_t_3);
+  __pyx_t_5 = 0;
   __pyx_t_6 = 0;
   __pyx_t_7 = 0;
-  __pyx_t_8 = 0;
-  __pyx_t_1 = 0;
-  __pyx_r = ((PyObject *)__pyx_t_2);
-  __pyx_t_2 = 0;
+  __pyx_t_3 = 0;
+  __pyx_r = ((PyObject *)__pyx_t_4);
+  __pyx_t_4 = 0;
   goto __pyx_L0;
 
   __pyx_r = Py_None; __Pyx_INCREF(Py_None);
   goto __pyx_L0;
   __pyx_L1_error:;
-  __Pyx_XDECREF(__pyx_t_1);
-  __Pyx_XDECREF(__pyx_t_2);
+  __Pyx_XDECREF(__pyx_t_3);
+  __Pyx_XDECREF(__pyx_t_4);
+  __Pyx_XDECREF(__pyx_t_5);
   __Pyx_XDECREF(__pyx_t_6);
   __Pyx_XDECREF(__pyx_t_7);
-  __Pyx_XDECREF(__pyx_t_8);
   { PyObject *__pyx_type, *__pyx_value, *__pyx_tb;
     __Pyx_ErrFetch(&__pyx_type, &__pyx_value, &__pyx_tb);
     __Pyx_SafeReleaseBuffer(&__pyx_pybuffernd_X.rcbuffer->pybuffer);
@@ -9020,58 +10360,12 @@ static PyObject *__pyx_tp_new_7sklearn_4tree_5_tree_ClassificationCriterion(PyTy
   if (!o) return 0;
   p = ((struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *)o);
   p->__pyx_base.__pyx_vtab = (struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion*)__pyx_vtabptr_7sklearn_4tree_5_tree_ClassificationCriterion;
-  p->ndarray_label_count_left = Py_None; Py_INCREF(Py_None);
-  p->ndarray_label_count_right = Py_None; Py_INCREF(Py_None);
-  p->ndarray_label_count_init = Py_None; Py_INCREF(Py_None);
   return o;
 }
 
-static void __pyx_tp_dealloc_7sklearn_4tree_5_tree_ClassificationCriterion(PyObject *o) {
-  struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *p = (struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *)o;
-  Py_XDECREF(p->ndarray_label_count_left);
-  Py_XDECREF(p->ndarray_label_count_right);
-  Py_XDECREF(p->ndarray_label_count_init);
-  __pyx_tp_dealloc_7sklearn_4tree_5_tree_Criterion(o);
-}
-
-static int __pyx_tp_traverse_7sklearn_4tree_5_tree_ClassificationCriterion(PyObject *o, visitproc v, void *a) {
-  int e;
-  struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *p = (struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *)o;
-  if (__pyx_ptype_7sklearn_4tree_5_tree_Criterion->tp_traverse) {
-    e = __pyx_ptype_7sklearn_4tree_5_tree_Criterion->tp_traverse(o, v, a); if (e) return e;
-  }
-  if (p->ndarray_label_count_left) {
-    e = (*v)(p->ndarray_label_count_left, a); if (e) return e;
-  }
-  if (p->ndarray_label_count_right) {
-    e = (*v)(p->ndarray_label_count_right, a); if (e) return e;
-  }
-  if (p->ndarray_label_count_init) {
-    e = (*v)(p->ndarray_label_count_init, a); if (e) return e;
-  }
-  return 0;
-}
-
-static int __pyx_tp_clear_7sklearn_4tree_5_tree_ClassificationCriterion(PyObject *o) {
-  struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *p = (struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion *)o;
-  PyObject* tmp;
-  if (__pyx_ptype_7sklearn_4tree_5_tree_Criterion->tp_clear) {
-    __pyx_ptype_7sklearn_4tree_5_tree_Criterion->tp_clear(o);
-  }
-  tmp = ((PyObject*)p->ndarray_label_count_left);
-  p->ndarray_label_count_left = Py_None; Py_INCREF(Py_None);
-  Py_XDECREF(tmp);
-  tmp = ((PyObject*)p->ndarray_label_count_right);
-  p->ndarray_label_count_right = Py_None; Py_INCREF(Py_None);
-  Py_XDECREF(tmp);
-  tmp = ((PyObject*)p->ndarray_label_count_init);
-  p->ndarray_label_count_init = Py_None; Py_INCREF(Py_None);
-  Py_XDECREF(tmp);
-  return 0;
-}
-
 static PyMethodDef __pyx_methods_7sklearn_4tree_5_tree_ClassificationCriterion[] = {
-  {__Pyx_NAMESTR("init_value"), (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_3init_value, METH_NOARGS, __Pyx_DOCSTR(0)},
+  {__Pyx_NAMESTR("__del__"), (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_3__del__, METH_NOARGS, __Pyx_DOCSTR(0)},
+  {__Pyx_NAMESTR("init_value"), (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_23ClassificationCriterion_5init_value, METH_NOARGS, __Pyx_DOCSTR(0)},
   {0, 0, 0, 0}
 };
 
@@ -9178,7 +10472,7 @@ static PyTypeObject __pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion = {
   __Pyx_NAMESTR("sklearn.tree._tree.ClassificationCriterion"), /*tp_name*/
   sizeof(struct __pyx_obj_7sklearn_4tree_5_tree_ClassificationCriterion), /*tp_basicsize*/
   0, /*tp_itemsize*/
-  __pyx_tp_dealloc_7sklearn_4tree_5_tree_ClassificationCriterion, /*tp_dealloc*/
+  __pyx_tp_dealloc_7sklearn_4tree_5_tree_Criterion, /*tp_dealloc*/
   0, /*tp_print*/
   0, /*tp_getattr*/
   0, /*tp_setattr*/
@@ -9197,10 +10491,10 @@ static PyTypeObject __pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion = {
   0, /*tp_getattro*/
   0, /*tp_setattro*/
   &__pyx_tp_as_buffer_ClassificationCriterion, /*tp_as_buffer*/
-  Py_TPFLAGS_DEFAULT|Py_TPFLAGS_CHECKTYPES|Py_TPFLAGS_HAVE_NEWBUFFER|Py_TPFLAGS_BASETYPE|Py_TPFLAGS_HAVE_GC, /*tp_flags*/
+  Py_TPFLAGS_DEFAULT|Py_TPFLAGS_CHECKTYPES|Py_TPFLAGS_HAVE_NEWBUFFER|Py_TPFLAGS_BASETYPE, /*tp_flags*/
   __Pyx_DOCSTR("Abstract criterion for classification.\n\n    Attributes\n    ----------\n    n_classes : int\n        The number of classes.\n\n    n_samples : int\n        The number of samples.\n\n    label_count_left : int*\n        The label counts for samples left of splitting point.\n\n    label_count_right : int*\n        The label counts for samples right of splitting point.\n\n    label_count_init : int*\n        The initial label counts for samples right of splitting point.\n        Used to reset `label_count_right` for each feature.\n\n    n_left : int\n        The number of samples left of splitting point.\n\n    n_right : int\n        The number of samples right of splitting point.\n    "), /*tp_doc*/
-  __pyx_tp_traverse_7sklearn_4tree_5_tree_ClassificationCriterion, /*tp_traverse*/
-  __pyx_tp_clear_7sklearn_4tree_5_tree_ClassificationCriterion, /*tp_clear*/
+  0, /*tp_traverse*/
+  0, /*tp_clear*/
   0, /*tp_richcompare*/
   0, /*tp_weaklistoffset*/
   0, /*tp_iter*/
@@ -9232,7 +10526,7 @@ static struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Gini __pyx_vtable_7sklearn_
 
 static PyObject *__pyx_tp_new_7sklearn_4tree_5_tree_Gini(PyTypeObject *t, PyObject *a, PyObject *k) {
   struct __pyx_obj_7sklearn_4tree_5_tree_Gini *p;
-  PyObject *o = __pyx_tp_new_7sklearn_4tree_5_tree_ClassificationCriterion(t, a, k);
+  PyObject *o = __pyx_tp_new_7sklearn_4tree_5_tree_Criterion(t, a, k);
   if (!o) return 0;
   p = ((struct __pyx_obj_7sklearn_4tree_5_tree_Gini *)o);
   p->__pyx_base.__pyx_base.__pyx_vtab = (struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion*)__pyx_vtabptr_7sklearn_4tree_5_tree_Gini;
@@ -9346,7 +10640,7 @@ static PyTypeObject __pyx_type_7sklearn_4tree_5_tree_Gini = {
   __Pyx_NAMESTR("sklearn.tree._tree.Gini"), /*tp_name*/
   sizeof(struct __pyx_obj_7sklearn_4tree_5_tree_Gini), /*tp_basicsize*/
   0, /*tp_itemsize*/
-  __pyx_tp_dealloc_7sklearn_4tree_5_tree_ClassificationCriterion, /*tp_dealloc*/
+  __pyx_tp_dealloc_7sklearn_4tree_5_tree_Criterion, /*tp_dealloc*/
   0, /*tp_print*/
   0, /*tp_getattr*/
   0, /*tp_setattr*/
@@ -9365,10 +10659,10 @@ static PyTypeObject __pyx_type_7sklearn_4tree_5_tree_Gini = {
   0, /*tp_getattro*/
   0, /*tp_setattro*/
   &__pyx_tp_as_buffer_Gini, /*tp_as_buffer*/
-  Py_TPFLAGS_DEFAULT|Py_TPFLAGS_CHECKTYPES|Py_TPFLAGS_HAVE_NEWBUFFER|Py_TPFLAGS_BASETYPE|Py_TPFLAGS_HAVE_GC, /*tp_flags*/
+  Py_TPFLAGS_DEFAULT|Py_TPFLAGS_CHECKTYPES|Py_TPFLAGS_HAVE_NEWBUFFER|Py_TPFLAGS_BASETYPE, /*tp_flags*/
   __Pyx_DOCSTR("Gini Index splitting criteria.\n\n    Gini index = \\sum_{k=0}^{K-1} pmk (1 - pmk)\n               = 1 - \\sum_{k=0}^{K-1} pmk ** 2\n    "), /*tp_doc*/
-  __pyx_tp_traverse_7sklearn_4tree_5_tree_ClassificationCriterion, /*tp_traverse*/
-  __pyx_tp_clear_7sklearn_4tree_5_tree_ClassificationCriterion, /*tp_clear*/
+  0, /*tp_traverse*/
+  0, /*tp_clear*/
   0, /*tp_richcompare*/
   0, /*tp_weaklistoffset*/
   0, /*tp_iter*/
@@ -9400,7 +10694,7 @@ static struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Entropy __pyx_vtable_7sklea
 
 static PyObject *__pyx_tp_new_7sklearn_4tree_5_tree_Entropy(PyTypeObject *t, PyObject *a, PyObject *k) {
   struct __pyx_obj_7sklearn_4tree_5_tree_Entropy *p;
-  PyObject *o = __pyx_tp_new_7sklearn_4tree_5_tree_ClassificationCriterion(t, a, k);
+  PyObject *o = __pyx_tp_new_7sklearn_4tree_5_tree_Criterion(t, a, k);
   if (!o) return 0;
   p = ((struct __pyx_obj_7sklearn_4tree_5_tree_Entropy *)o);
   p->__pyx_base.__pyx_base.__pyx_vtab = (struct __pyx_vtabstruct_7sklearn_4tree_5_tree_Criterion*)__pyx_vtabptr_7sklearn_4tree_5_tree_Entropy;
@@ -9514,7 +10808,7 @@ static PyTypeObject __pyx_type_7sklearn_4tree_5_tree_Entropy = {
   __Pyx_NAMESTR("sklearn.tree._tree.Entropy"), /*tp_name*/
   sizeof(struct __pyx_obj_7sklearn_4tree_5_tree_Entropy), /*tp_basicsize*/
   0, /*tp_itemsize*/
-  __pyx_tp_dealloc_7sklearn_4tree_5_tree_ClassificationCriterion, /*tp_dealloc*/
+  __pyx_tp_dealloc_7sklearn_4tree_5_tree_Criterion, /*tp_dealloc*/
   0, /*tp_print*/
   0, /*tp_getattr*/
   0, /*tp_setattr*/
@@ -9533,10 +10827,10 @@ static PyTypeObject __pyx_type_7sklearn_4tree_5_tree_Entropy = {
   0, /*tp_getattro*/
   0, /*tp_setattro*/
   &__pyx_tp_as_buffer_Entropy, /*tp_as_buffer*/
-  Py_TPFLAGS_DEFAULT|Py_TPFLAGS_CHECKTYPES|Py_TPFLAGS_HAVE_NEWBUFFER|Py_TPFLAGS_BASETYPE|Py_TPFLAGS_HAVE_GC, /*tp_flags*/
+  Py_TPFLAGS_DEFAULT|Py_TPFLAGS_CHECKTYPES|Py_TPFLAGS_HAVE_NEWBUFFER|Py_TPFLAGS_BASETYPE, /*tp_flags*/
   __Pyx_DOCSTR("Entropy splitting criteria.\n\n    Cross Entropy = - \\sum_{k=0}^{K-1} pmk log(pmk)\n    "), /*tp_doc*/
-  __pyx_tp_traverse_7sklearn_4tree_5_tree_ClassificationCriterion, /*tp_traverse*/
-  __pyx_tp_clear_7sklearn_4tree_5_tree_ClassificationCriterion, /*tp_clear*/
+  0, /*tp_traverse*/
+  0, /*tp_clear*/
   0, /*tp_richcompare*/
   0, /*tp_weaklistoffset*/
   0, /*tp_iter*/
@@ -9576,7 +10870,8 @@ static PyObject *__pyx_tp_new_7sklearn_4tree_5_tree_RegressionCriterion(PyTypeOb
 }
 
 static PyMethodDef __pyx_methods_7sklearn_4tree_5_tree_RegressionCriterion[] = {
-  {__Pyx_NAMESTR("init_value"), (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_3init_value, METH_NOARGS, __Pyx_DOCSTR(0)},
+  {__Pyx_NAMESTR("__del__"), (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_3__del__, METH_NOARGS, __Pyx_DOCSTR(0)},
+  {__Pyx_NAMESTR("init_value"), (PyCFunction)__pyx_pw_7sklearn_4tree_5_tree_19RegressionCriterion_5init_value, METH_NOARGS, __Pyx_DOCSTR(0)},
   {0, 0, 0, 0}
 };
 
@@ -9932,9 +11227,7 @@ static __Pyx_StringTabEntry __pyx_string_tab[] = {
   {&__pyx_kp_u_5, __pyx_k_5, sizeof(__pyx_k_5), 0, 1, 0, 0},
   {&__pyx_kp_u_7, __pyx_k_7, sizeof(__pyx_k_7), 0, 1, 0, 0},
   {&__pyx_kp_u_8, __pyx_k_8, sizeof(__pyx_k_8), 0, 1, 0, 0},
-  {&__pyx_n_s__C, __pyx_k__C, sizeof(__pyx_k__C), 0, 0, 1, 1},
   {&__pyx_n_s__DTYPE, __pyx_k__DTYPE, sizeof(__pyx_k__DTYPE), 0, 0, 1, 1},
-  {&__pyx_n_s__K, __pyx_k__K, sizeof(__pyx_k__K), 0, 0, 1, 1},
   {&__pyx_n_s__RuntimeError, __pyx_k__RuntimeError, sizeof(__pyx_k__RuntimeError), 0, 0, 1, 1},
   {&__pyx_n_s__ValueError, __pyx_k__ValueError, sizeof(__pyx_k__ValueError), 0, 0, 1, 1},
   {&__pyx_n_s__X, __pyx_k__X, sizeof(__pyx_k__X), 0, 0, 1, 1},
@@ -9954,7 +11247,6 @@ static __Pyx_StringTabEntry __pyx_string_tab[] = {
   {&__pyx_n_s___random_sample_mask, __pyx_k___random_sample_mask, sizeof(__pyx_k___random_sample_mask), 0, 0, 1, 1},
   {&__pyx_n_s__a, __pyx_k__a, sizeof(__pyx_k__a), 0, 0, 1, 1},
   {&__pyx_n_s__arange, __pyx_k__arange, sizeof(__pyx_k__arange), 0, 0, 1, 1},
-  {&__pyx_n_s__asarray, __pyx_k__asarray, sizeof(__pyx_k__asarray), 0, 0, 1, 1},
   {&__pyx_n_s__astype, __pyx_k__astype, sizeof(__pyx_k__astype), 0, 0, 1, 1},
   {&__pyx_n_s__b, __pyx_k__b, sizeof(__pyx_k__b), 0, 0, 1, 1},
   {&__pyx_n_s__best_error, __pyx_k__best_error, sizeof(__pyx_k__best_error), 0, 0, 1, 1},
@@ -9984,13 +11276,13 @@ static __Pyx_StringTabEntry __pyx_string_tab[] = {
   {&__pyx_n_s__n_classes, __pyx_k__n_classes, sizeof(__pyx_k__n_classes), 0, 0, 1, 1},
   {&__pyx_n_s__n_features, __pyx_k__n_features, sizeof(__pyx_k__n_features), 0, 0, 1, 1},
   {&__pyx_n_s__n_left, __pyx_k__n_left, sizeof(__pyx_k__n_left), 0, 0, 1, 1},
+  {&__pyx_n_s__n_outputs, __pyx_k__n_outputs, sizeof(__pyx_k__n_outputs), 0, 0, 1, 1},
   {&__pyx_n_s__n_samples, __pyx_k__n_samples, sizeof(__pyx_k__n_samples), 0, 0, 1, 1},
   {&__pyx_n_s__n_total_in_bag, __pyx_k__n_total_in_bag, sizeof(__pyx_k__n_total_in_bag), 0, 0, 1, 1},
   {&__pyx_n_s__n_total_samples, __pyx_k__n_total_samples, sizeof(__pyx_k__n_total_samples), 0, 0, 1, 1},
   {&__pyx_n_s__node_id, __pyx_k__node_id, sizeof(__pyx_k__node_id), 0, 0, 1, 1},
   {&__pyx_n_s__np, __pyx_k__np, sizeof(__pyx_k__np), 0, 0, 1, 1},
   {&__pyx_n_s__numpy, __pyx_k__numpy, sizeof(__pyx_k__numpy), 0, 0, 1, 1},
-  {&__pyx_n_s__order, __pyx_k__order, sizeof(__pyx_k__order), 0, 0, 1, 1},
   {&__pyx_n_s__out, __pyx_k__out, sizeof(__pyx_k__out), 0, 0, 1, 1},
   {&__pyx_n_s__permutation, __pyx_k__permutation, sizeof(__pyx_k__permutation), 0, 0, 1, 1},
   {&__pyx_n_s__pred, __pyx_k__pred, sizeof(__pyx_k__pred), 0, 0, 1, 1},
@@ -10002,20 +11294,15 @@ static __Pyx_StringTabEntry __pyx_string_tab[] = {
   {&__pyx_n_s__t, __pyx_k__t, sizeof(__pyx_k__t), 0, 0, 1, 1},
   {&__pyx_n_s__threshold, __pyx_k__threshold, sizeof(__pyx_k__threshold), 0, 0, 1, 1},
   {&__pyx_n_s__values, __pyx_k__values, sizeof(__pyx_k__values), 0, 0, 1, 1},
-  {&__pyx_n_s__xrange, __pyx_k__xrange, sizeof(__pyx_k__xrange), 0, 0, 1, 1},
   {&__pyx_n_s__y, __pyx_k__y, sizeof(__pyx_k__y), 0, 0, 1, 1},
   {&__pyx_n_s__y_ptr, __pyx_k__y_ptr, sizeof(__pyx_k__y_ptr), 0, 0, 1, 1},
+  {&__pyx_n_s__y_stride, __pyx_k__y_stride, sizeof(__pyx_k__y_stride), 0, 0, 1, 1},
   {&__pyx_n_s__zeros, __pyx_k__zeros, sizeof(__pyx_k__zeros), 0, 0, 1, 1},
   {0, 0, 0, 0, 0, 0, 0}
 };
 static int __Pyx_InitCachedBuiltins(void) {
-  __pyx_builtin_range = __Pyx_GetName(__pyx_b, __pyx_n_s__range); if (!__pyx_builtin_range) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 434; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  #if PY_MAJOR_VERSION >= 3
-  __pyx_builtin_xrange = __Pyx_GetName(__pyx_b, __pyx_n_s__range); if (!__pyx_builtin_xrange) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 452; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  #else
-  __pyx_builtin_xrange = __Pyx_GetName(__pyx_b, __pyx_n_s__xrange); if (!__pyx_builtin_xrange) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 452; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  #endif
   __pyx_builtin_ValueError = __Pyx_GetName(__pyx_b, __pyx_n_s__ValueError); if (!__pyx_builtin_ValueError) {__pyx_filename = __pyx_f[1]; __pyx_lineno = 214; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_builtin_range = __Pyx_GetName(__pyx_b, __pyx_n_s__range); if (!__pyx_builtin_range) {__pyx_filename = __pyx_f[1]; __pyx_lineno = 227; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_builtin_RuntimeError = __Pyx_GetName(__pyx_b, __pyx_n_s__RuntimeError); if (!__pyx_builtin_RuntimeError) {__pyx_filename = __pyx_f[1]; __pyx_lineno = 798; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   return 0;
   __pyx_L1_error:;
@@ -10110,14 +11397,14 @@ static int __Pyx_InitCachedConstants(void) {
   __Pyx_GIVEREF(((PyObject *)__pyx_kp_u_11));
   __Pyx_GIVEREF(((PyObject *)__pyx_k_tuple_12));
 
-  /* "sklearn/tree/_tree.pyx":409
+  /* "sklearn/tree/_tree.pyx":604
  * 
  * 
  * def _random_sample_mask(int n_total_samples, int n_total_in_bag, random_state):             # <<<<<<<<<<<<<<
  *     """Create a random sample mask where ``n_total_in_bag`` elements are set.
  * 
  */
-  __pyx_k_tuple_13 = PyTuple_New(7); if (unlikely(!__pyx_k_tuple_13)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_tuple_13 = PyTuple_New(7); if (unlikely(!__pyx_k_tuple_13)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_k_tuple_13);
   __Pyx_INCREF(((PyObject *)__pyx_n_s__n_total_samples));
   PyTuple_SET_ITEM(__pyx_k_tuple_13, 0, ((PyObject *)__pyx_n_s__n_total_samples));
@@ -10141,16 +11428,16 @@ static int __Pyx_InitCachedConstants(void) {
   PyTuple_SET_ITEM(__pyx_k_tuple_13, 6, ((PyObject *)__pyx_n_s__i));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__i));
   __Pyx_GIVEREF(((PyObject *)__pyx_k_tuple_13));
-  __pyx_k_codeobj_14 = (PyObject*)__Pyx_PyCode_New(3, 0, 7, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_13, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___random_sample_mask, 409, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_14)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_codeobj_14 = (PyObject*)__Pyx_PyCode_New(3, 0, 7, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_13, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___random_sample_mask, 604, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_14)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
 
-  /* "sklearn/tree/_tree.pyx":442
+  /* "sklearn/tree/_tree.pyx":637
  * 
  * 
  * def _apply_tree(np.ndarray[DTYPE_t, ndim=2] X,             # <<<<<<<<<<<<<<
  *                 np.ndarray[np.int32_t, ndim=2] children,
  *                 np.ndarray[np.int32_t, ndim=1] feature,
  */
-  __pyx_k_tuple_17 = PyTuple_New(8); if (unlikely(!__pyx_k_tuple_17)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_tuple_17 = PyTuple_New(8); if (unlikely(!__pyx_k_tuple_17)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_k_tuple_17);
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X));
   PyTuple_SET_ITEM(__pyx_k_tuple_17, 0, ((PyObject *)__pyx_n_s__X));
@@ -10177,16 +11464,16 @@ static int __Pyx_InitCachedConstants(void) {
   PyTuple_SET_ITEM(__pyx_k_tuple_17, 7, ((PyObject *)__pyx_n_s__node_id));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__node_id));
   __Pyx_GIVEREF(((PyObject *)__pyx_k_tuple_17));
-  __pyx_k_codeobj_18 = (PyObject*)__Pyx_PyCode_New(5, 0, 8, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_17, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___apply_tree, 442, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_18)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_codeobj_18 = (PyObject*)__Pyx_PyCode_New(5, 0, 8, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_17, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___apply_tree, 637, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_18)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
 
-  /* "sklearn/tree/_tree.pyx":463
+  /* "sklearn/tree/_tree.pyx":658
  * 
  * 
  * def _predict_tree(np.ndarray[DTYPE_t, ndim=2] X,             # <<<<<<<<<<<<<<
  *                   np.ndarray[np.int32_t, ndim=2] children,
  *                   np.ndarray[np.int32_t, ndim=1] feature,
  */
-  __pyx_k_tuple_19 = PyTuple_New(11); if (unlikely(!__pyx_k_tuple_19)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_tuple_19 = PyTuple_New(13); if (unlikely(!__pyx_k_tuple_19)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_k_tuple_19);
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X));
   PyTuple_SET_ITEM(__pyx_k_tuple_19, 0, ((PyObject *)__pyx_n_s__X));
@@ -10209,29 +11496,35 @@ static int __Pyx_InitCachedConstants(void) {
   __Pyx_INCREF(((PyObject *)__pyx_n_s__i));
   PyTuple_SET_ITEM(__pyx_k_tuple_19, 6, ((PyObject *)__pyx_n_s__i));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__i));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__k));
+  PyTuple_SET_ITEM(__pyx_k_tuple_19, 7, ((PyObject *)__pyx_n_s__k));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__k));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__c));
+  PyTuple_SET_ITEM(__pyx_k_tuple_19, 8, ((PyObject *)__pyx_n_s__c));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__c));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__n));
-  PyTuple_SET_ITEM(__pyx_k_tuple_19, 7, ((PyObject *)__pyx_n_s__n));
+  PyTuple_SET_ITEM(__pyx_k_tuple_19, 9, ((PyObject *)__pyx_n_s__n));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__n));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__node_id));
-  PyTuple_SET_ITEM(__pyx_k_tuple_19, 8, ((PyObject *)__pyx_n_s__node_id));
+  PyTuple_SET_ITEM(__pyx_k_tuple_19, 10, ((PyObject *)__pyx_n_s__node_id));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__node_id));
-  __Pyx_INCREF(((PyObject *)__pyx_n_s__K));
-  PyTuple_SET_ITEM(__pyx_k_tuple_19, 9, ((PyObject *)__pyx_n_s__K));
-  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__K));
-  __Pyx_INCREF(((PyObject *)__pyx_n_s__k));
-  PyTuple_SET_ITEM(__pyx_k_tuple_19, 10, ((PyObject *)__pyx_n_s__k));
-  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__k));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__n_outputs));
+  PyTuple_SET_ITEM(__pyx_k_tuple_19, 11, ((PyObject *)__pyx_n_s__n_outputs));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__n_outputs));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__n_classes));
+  PyTuple_SET_ITEM(__pyx_k_tuple_19, 12, ((PyObject *)__pyx_n_s__n_classes));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__n_classes));
   __Pyx_GIVEREF(((PyObject *)__pyx_k_tuple_19));
-  __pyx_k_codeobj_20 = (PyObject*)__Pyx_PyCode_New(6, 0, 11, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_19, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___predict_tree, 463, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_20)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_codeobj_20 = (PyObject*)__Pyx_PyCode_New(6, 0, 13, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_19, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___predict_tree, 658, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_20)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
 
-  /* "sklearn/tree/_tree.pyx":486
+  /* "sklearn/tree/_tree.pyx":685
  * 
  * 
- * def _error_at_leaf(np.ndarray[DTYPE_t, ndim=1, mode="c"] y,             # <<<<<<<<<<<<<<
- *                    np.ndarray sample_mask, Criterion criterion,
- *                    int n_samples):
+ * def _error_at_leaf(np.ndarray[DTYPE_t, ndim=2, mode="c"] y,             # <<<<<<<<<<<<<<
+ *                    np.ndarray sample_mask,
+ *                    Criterion criterion,
  */
-  __pyx_k_tuple_21 = PyTuple_New(7); if (unlikely(!__pyx_k_tuple_21)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_tuple_21 = PyTuple_New(8); if (unlikely(!__pyx_k_tuple_21)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_k_tuple_21);
   __Pyx_INCREF(((PyObject *)__pyx_n_s__y));
   PyTuple_SET_ITEM(__pyx_k_tuple_21, 0, ((PyObject *)__pyx_n_s__y));
@@ -10245,26 +11538,29 @@ static int __Pyx_InitCachedConstants(void) {
   __Pyx_INCREF(((PyObject *)__pyx_n_s__n_samples));
   PyTuple_SET_ITEM(__pyx_k_tuple_21, 3, ((PyObject *)__pyx_n_s__n_samples));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__n_samples));
-  __Pyx_INCREF(((PyObject *)__pyx_n_s__n_total_samples));
-  PyTuple_SET_ITEM(__pyx_k_tuple_21, 4, ((PyObject *)__pyx_n_s__n_total_samples));
-  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__n_total_samples));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__y_ptr));
-  PyTuple_SET_ITEM(__pyx_k_tuple_21, 5, ((PyObject *)__pyx_n_s__y_ptr));
+  PyTuple_SET_ITEM(__pyx_k_tuple_21, 4, ((PyObject *)__pyx_n_s__y_ptr));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__y_ptr));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__y_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_21, 5, ((PyObject *)__pyx_n_s__y_stride));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__y_stride));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__n_total_samples));
+  PyTuple_SET_ITEM(__pyx_k_tuple_21, 6, ((PyObject *)__pyx_n_s__n_total_samples));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__n_total_samples));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__sample_mask_ptr));
-  PyTuple_SET_ITEM(__pyx_k_tuple_21, 6, ((PyObject *)__pyx_n_s__sample_mask_ptr));
+  PyTuple_SET_ITEM(__pyx_k_tuple_21, 7, ((PyObject *)__pyx_n_s__sample_mask_ptr));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__sample_mask_ptr));
   __Pyx_GIVEREF(((PyObject *)__pyx_k_tuple_21));
-  __pyx_k_codeobj_22 = (PyObject*)__Pyx_PyCode_New(4, 0, 7, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_21, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___error_at_leaf, 486, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_22)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_codeobj_22 = (PyObject*)__Pyx_PyCode_New(4, 0, 8, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_21, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___error_at_leaf, 685, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_22)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
 
-  /* "sklearn/tree/_tree.pyx":533
+  /* "sklearn/tree/_tree.pyx":736
  * 
  * 
  * def _find_best_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,             # <<<<<<<<<<<<<<
- *                      np.ndarray[DTYPE_t, ndim=1, mode="c"] y,
+ *                      np.ndarray[DTYPE_t, ndim=2, mode="c"] y,
  *                      np.ndarray[np.int32_t, ndim=2, mode="fortran"] X_argsorted,
  */
-  __pyx_k_tuple_25 = PyTuple_New(33); if (unlikely(!__pyx_k_tuple_25)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_tuple_25 = PyTuple_New(34); if (unlikely(!__pyx_k_tuple_25)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_k_tuple_25);
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X));
   PyTuple_SET_ITEM(__pyx_k_tuple_25, 0, ((PyObject *)__pyx_n_s__X));
@@ -10332,50 +11628,53 @@ static int __Pyx_InitCachedConstants(void) {
   __Pyx_INCREF(((PyObject *)__pyx_n_s__best_t));
   PyTuple_SET_ITEM(__pyx_k_tuple_25, 21, ((PyObject *)__pyx_n_s__best_t));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__best_t));
-  __Pyx_INCREF(((PyObject *)__pyx_n_s__y_ptr));
-  PyTuple_SET_ITEM(__pyx_k_tuple_25, 22, ((PyObject *)__pyx_n_s__y_ptr));
-  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__y_ptr));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_i));
-  PyTuple_SET_ITEM(__pyx_k_tuple_25, 23, ((PyObject *)__pyx_n_s__X_i));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 22, ((PyObject *)__pyx_n_s__X_i));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_i));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_argsorted_i));
-  PyTuple_SET_ITEM(__pyx_k_tuple_25, 24, ((PyObject *)__pyx_n_s__X_argsorted_i));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 23, ((PyObject *)__pyx_n_s__X_argsorted_i));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_argsorted_i));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__y_ptr));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 24, ((PyObject *)__pyx_n_s__y_ptr));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__y_ptr));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__sample_mask_ptr));
   PyTuple_SET_ITEM(__pyx_k_tuple_25, 25, ((PyObject *)__pyx_n_s__sample_mask_ptr));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__sample_mask_ptr));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__features));
   PyTuple_SET_ITEM(__pyx_k_tuple_25, 26, ((PyObject *)__pyx_n_s__features));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__features));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__y_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 27, ((PyObject *)__pyx_n_s__y_stride));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__y_stride));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_elem_stride));
-  PyTuple_SET_ITEM(__pyx_k_tuple_25, 27, ((PyObject *)__pyx_n_s__X_elem_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 28, ((PyObject *)__pyx_n_s__X_elem_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_elem_stride));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_col_stride));
-  PyTuple_SET_ITEM(__pyx_k_tuple_25, 28, ((PyObject *)__pyx_n_s__X_col_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 29, ((PyObject *)__pyx_n_s__X_col_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_col_stride));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_stride));
-  PyTuple_SET_ITEM(__pyx_k_tuple_25, 29, ((PyObject *)__pyx_n_s__X_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 30, ((PyObject *)__pyx_n_s__X_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_stride));
   __Pyx_INCREF(((PyObject *)__pyx_n_s_23));
-  PyTuple_SET_ITEM(__pyx_k_tuple_25, 30, ((PyObject *)__pyx_n_s_23));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 31, ((PyObject *)__pyx_n_s_23));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s_23));
   __Pyx_INCREF(((PyObject *)__pyx_n_s_24));
-  PyTuple_SET_ITEM(__pyx_k_tuple_25, 31, ((PyObject *)__pyx_n_s_24));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 32, ((PyObject *)__pyx_n_s_24));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s_24));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_argsorted_stride));
-  PyTuple_SET_ITEM(__pyx_k_tuple_25, 32, ((PyObject *)__pyx_n_s__X_argsorted_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_25, 33, ((PyObject *)__pyx_n_s__X_argsorted_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_argsorted_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_k_tuple_25));
-  __pyx_k_codeobj_26 = (PyObject*)__Pyx_PyCode_New(9, 0, 33, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_25, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___find_best_split, 533, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_26)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_codeobj_26 = (PyObject*)__Pyx_PyCode_New(9, 0, 34, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_25, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s___find_best_split, 736, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_26)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
 
-  /* "sklearn/tree/_tree.pyx":678
+  /* "sklearn/tree/_tree.pyx":882
  *     return best_i, best_t, best_error, initial_error
  * 
  * def _find_best_random_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,             # <<<<<<<<<<<<<<
- *                             np.ndarray[DTYPE_t, ndim=1, mode="c"] y,
+ *                             np.ndarray[DTYPE_t, ndim=2, mode="c"] y,
  *                             np.ndarray[np.int32_t, ndim=2, mode="fortran"] X_argsorted,
  */
-  __pyx_k_tuple_27 = PyTuple_New(34); if (unlikely(!__pyx_k_tuple_27)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_tuple_27 = PyTuple_New(35); if (unlikely(!__pyx_k_tuple_27)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_k_tuple_27);
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X));
   PyTuple_SET_ITEM(__pyx_k_tuple_27, 0, ((PyObject *)__pyx_n_s__X));
@@ -10446,41 +11745,44 @@ static int __Pyx_InitCachedConstants(void) {
   __Pyx_INCREF(((PyObject *)__pyx_n_s__best_t));
   PyTuple_SET_ITEM(__pyx_k_tuple_27, 22, ((PyObject *)__pyx_n_s__best_t));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__best_t));
-  __Pyx_INCREF(((PyObject *)__pyx_n_s__y_ptr));
-  PyTuple_SET_ITEM(__pyx_k_tuple_27, 23, ((PyObject *)__pyx_n_s__y_ptr));
-  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__y_ptr));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_i));
-  PyTuple_SET_ITEM(__pyx_k_tuple_27, 24, ((PyObject *)__pyx_n_s__X_i));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 23, ((PyObject *)__pyx_n_s__X_i));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_i));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_argsorted_i));
-  PyTuple_SET_ITEM(__pyx_k_tuple_27, 25, ((PyObject *)__pyx_n_s__X_argsorted_i));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 24, ((PyObject *)__pyx_n_s__X_argsorted_i));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_argsorted_i));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__y_ptr));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 25, ((PyObject *)__pyx_n_s__y_ptr));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__y_ptr));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__sample_mask_ptr));
   PyTuple_SET_ITEM(__pyx_k_tuple_27, 26, ((PyObject *)__pyx_n_s__sample_mask_ptr));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__sample_mask_ptr));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__features));
   PyTuple_SET_ITEM(__pyx_k_tuple_27, 27, ((PyObject *)__pyx_n_s__features));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__features));
+  __Pyx_INCREF(((PyObject *)__pyx_n_s__y_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 28, ((PyObject *)__pyx_n_s__y_stride));
+  __Pyx_GIVEREF(((PyObject *)__pyx_n_s__y_stride));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_elem_stride));
-  PyTuple_SET_ITEM(__pyx_k_tuple_27, 28, ((PyObject *)__pyx_n_s__X_elem_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 29, ((PyObject *)__pyx_n_s__X_elem_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_elem_stride));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_col_stride));
-  PyTuple_SET_ITEM(__pyx_k_tuple_27, 29, ((PyObject *)__pyx_n_s__X_col_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 30, ((PyObject *)__pyx_n_s__X_col_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_col_stride));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_stride));
-  PyTuple_SET_ITEM(__pyx_k_tuple_27, 30, ((PyObject *)__pyx_n_s__X_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 31, ((PyObject *)__pyx_n_s__X_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_stride));
   __Pyx_INCREF(((PyObject *)__pyx_n_s_23));
-  PyTuple_SET_ITEM(__pyx_k_tuple_27, 31, ((PyObject *)__pyx_n_s_23));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 32, ((PyObject *)__pyx_n_s_23));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s_23));
   __Pyx_INCREF(((PyObject *)__pyx_n_s_24));
-  PyTuple_SET_ITEM(__pyx_k_tuple_27, 32, ((PyObject *)__pyx_n_s_24));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 33, ((PyObject *)__pyx_n_s_24));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s_24));
   __Pyx_INCREF(((PyObject *)__pyx_n_s__X_argsorted_stride));
-  PyTuple_SET_ITEM(__pyx_k_tuple_27, 33, ((PyObject *)__pyx_n_s__X_argsorted_stride));
+  PyTuple_SET_ITEM(__pyx_k_tuple_27, 34, ((PyObject *)__pyx_n_s__X_argsorted_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_n_s__X_argsorted_stride));
   __Pyx_GIVEREF(((PyObject *)__pyx_k_tuple_27));
-  __pyx_k_codeobj_28 = (PyObject*)__Pyx_PyCode_New(9, 0, 34, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_27, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s_29, 678, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_28)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_k_codeobj_28 = (PyObject*)__Pyx_PyCode_New(9, 0, 35, 0, 0, __pyx_empty_bytes, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_k_tuple_27, __pyx_empty_tuple, __pyx_empty_tuple, __pyx_kp_s_15, __pyx_n_s_29, 882, __pyx_empty_bytes); if (unlikely(!__pyx_k_codeobj_28)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_RefNannyFinishContext();
   return 0;
   __pyx_L1_error:;
@@ -10490,6 +11792,7 @@ static int __Pyx_InitCachedConstants(void) {
 
 static int __Pyx_InitGlobals(void) {
   if (__Pyx_InitStrings(__pyx_string_tab) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1; __pyx_clineno = __LINE__; goto __pyx_L1_error;};
+  __pyx_int_1 = PyInt_FromLong(1); if (unlikely(!__pyx_int_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1; __pyx_clineno = __LINE__; goto __pyx_L1_error;};
   __pyx_int_15 = PyInt_FromLong(15); if (unlikely(!__pyx_int_15)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1; __pyx_clineno = __LINE__; goto __pyx_L1_error;};
   return 0;
   __pyx_L1_error:;
@@ -10506,6 +11809,7 @@ PyMODINIT_FUNC PyInit__tree(void)
 {
   PyObject *__pyx_t_1 = NULL;
   PyObject *__pyx_t_2 = NULL;
+  __pyx_t_7sklearn_4tree_5_tree_DTYPE_t __pyx_t_3;
   __Pyx_RefNannyDeclarations
   #if CYTHON_REFNANNY
   __Pyx_RefNanny = __Pyx_RefNannyImportAPI("refnanny");
@@ -10563,62 +11867,62 @@ PyMODINIT_FUNC PyInit__tree(void)
   /*--- Function export code ---*/
   /*--- Type init code ---*/
   __pyx_vtabptr_7sklearn_4tree_5_tree_Criterion = &__pyx_vtable_7sklearn_4tree_5_tree_Criterion;
-  __pyx_vtable_7sklearn_4tree_5_tree_Criterion.init = (void (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *, int, int))__pyx_f_7sklearn_4tree_5_tree_9Criterion_init;
+  __pyx_vtable_7sklearn_4tree_5_tree_Criterion.init = (void (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *, int, int))__pyx_f_7sklearn_4tree_5_tree_9Criterion_init;
   __pyx_vtable_7sklearn_4tree_5_tree_Criterion.reset = (void (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *))__pyx_f_7sklearn_4tree_5_tree_9Criterion_reset;
-  __pyx_vtable_7sklearn_4tree_5_tree_Criterion.update = (int (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int, int, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *))__pyx_f_7sklearn_4tree_5_tree_9Criterion_update;
+  __pyx_vtable_7sklearn_4tree_5_tree_Criterion.update = (int (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int, int, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int, int *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *))__pyx_f_7sklearn_4tree_5_tree_9Criterion_update;
   __pyx_vtable_7sklearn_4tree_5_tree_Criterion.eval = (double (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *))__pyx_f_7sklearn_4tree_5_tree_9Criterion_eval;
   __pyx_vtable_7sklearn_4tree_5_tree_Criterion.init_value = (PyArrayObject *(*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int __pyx_skip_dispatch))__pyx_f_7sklearn_4tree_5_tree_9Criterion_init_value;
-  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_Criterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 40; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_Criterion.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_Criterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 40; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetAttrString(__pyx_m, "Criterion", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_Criterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 40; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_Criterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 47; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_Criterion.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_Criterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 47; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetAttrString(__pyx_m, "Criterion", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_Criterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 47; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_ptype_7sklearn_4tree_5_tree_Criterion = &__pyx_type_7sklearn_4tree_5_tree_Criterion;
   __pyx_vtabptr_7sklearn_4tree_5_tree_ClassificationCriterion = &__pyx_vtable_7sklearn_4tree_5_tree_ClassificationCriterion;
   __pyx_vtable_7sklearn_4tree_5_tree_ClassificationCriterion.__pyx_base = *__pyx_vtabptr_7sklearn_4tree_5_tree_Criterion;
-  __pyx_vtable_7sklearn_4tree_5_tree_ClassificationCriterion.__pyx_base.init = (void (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *, int, int))__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init;
+  __pyx_vtable_7sklearn_4tree_5_tree_ClassificationCriterion.__pyx_base.init = (void (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *, int, int))__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init;
   __pyx_vtable_7sklearn_4tree_5_tree_ClassificationCriterion.__pyx_base.reset = (void (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *))__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_reset;
-  __pyx_vtable_7sklearn_4tree_5_tree_ClassificationCriterion.__pyx_base.update = (int (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int, int, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *))__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_update;
+  __pyx_vtable_7sklearn_4tree_5_tree_ClassificationCriterion.__pyx_base.update = (int (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int, int, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int, int *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *))__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_update;
   __pyx_vtable_7sklearn_4tree_5_tree_ClassificationCriterion.__pyx_base.eval = (double (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *))__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_eval;
   __pyx_vtable_7sklearn_4tree_5_tree_ClassificationCriterion.__pyx_base.init_value = (PyArrayObject *(*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int __pyx_skip_dispatch))__pyx_f_7sklearn_4tree_5_tree_23ClassificationCriterion_init_value;
   __pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion.tp_base = __pyx_ptype_7sklearn_4tree_5_tree_Criterion;
-  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 67; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_ClassificationCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 67; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetAttrString(__pyx_m, "ClassificationCriterion", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 67; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 81; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_ClassificationCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 81; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetAttrString(__pyx_m, "ClassificationCriterion", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 81; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_ptype_7sklearn_4tree_5_tree_ClassificationCriterion = &__pyx_type_7sklearn_4tree_5_tree_ClassificationCriterion;
   __pyx_vtabptr_7sklearn_4tree_5_tree_Gini = &__pyx_vtable_7sklearn_4tree_5_tree_Gini;
   __pyx_vtable_7sklearn_4tree_5_tree_Gini.__pyx_base = *__pyx_vtabptr_7sklearn_4tree_5_tree_ClassificationCriterion;
   __pyx_vtable_7sklearn_4tree_5_tree_Gini.__pyx_base.__pyx_base.eval = (double (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *))__pyx_f_7sklearn_4tree_5_tree_4Gini_eval;
   __pyx_type_7sklearn_4tree_5_tree_Gini.tp_base = __pyx_ptype_7sklearn_4tree_5_tree_ClassificationCriterion;
-  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_Gini) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 181; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_Gini.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_Gini) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 181; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetAttrString(__pyx_m, "Gini", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_Gini) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 181; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_Gini) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 255; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_Gini.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_Gini) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 255; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetAttrString(__pyx_m, "Gini", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_Gini) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 255; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_ptype_7sklearn_4tree_5_tree_Gini = &__pyx_type_7sklearn_4tree_5_tree_Gini;
   __pyx_vtabptr_7sklearn_4tree_5_tree_Entropy = &__pyx_vtable_7sklearn_4tree_5_tree_Entropy;
   __pyx_vtable_7sklearn_4tree_5_tree_Entropy.__pyx_base = *__pyx_vtabptr_7sklearn_4tree_5_tree_ClassificationCriterion;
   __pyx_vtable_7sklearn_4tree_5_tree_Entropy.__pyx_base.__pyx_base.eval = (double (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *))__pyx_f_7sklearn_4tree_5_tree_7Entropy_eval;
   __pyx_type_7sklearn_4tree_5_tree_Entropy.tp_base = __pyx_ptype_7sklearn_4tree_5_tree_ClassificationCriterion;
-  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_Entropy) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 217; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_Entropy.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_Entropy) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 217; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetAttrString(__pyx_m, "Entropy", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_Entropy) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 217; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_Entropy) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 306; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_Entropy.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_Entropy) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 306; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetAttrString(__pyx_m, "Entropy", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_Entropy) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 306; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_ptype_7sklearn_4tree_5_tree_Entropy = &__pyx_type_7sklearn_4tree_5_tree_Entropy;
   __pyx_vtabptr_7sklearn_4tree_5_tree_RegressionCriterion = &__pyx_vtable_7sklearn_4tree_5_tree_RegressionCriterion;
   __pyx_vtable_7sklearn_4tree_5_tree_RegressionCriterion.__pyx_base = *__pyx_vtabptr_7sklearn_4tree_5_tree_Criterion;
-  __pyx_vtable_7sklearn_4tree_5_tree_RegressionCriterion.__pyx_base.init = (void (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *, int, int))__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init;
+  __pyx_vtable_7sklearn_4tree_5_tree_RegressionCriterion.__pyx_base.init = (void (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *, int, int))__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init;
   __pyx_vtable_7sklearn_4tree_5_tree_RegressionCriterion.__pyx_base.reset = (void (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *))__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_reset;
-  __pyx_vtable_7sklearn_4tree_5_tree_RegressionCriterion.__pyx_base.update = (int (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int, int, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *))__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_update;
+  __pyx_vtable_7sklearn_4tree_5_tree_RegressionCriterion.__pyx_base.update = (int (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int, int, __pyx_t_7sklearn_4tree_5_tree_DTYPE_t *, int, int *, __pyx_t_7sklearn_4tree_5_tree_BOOL_t *))__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_update;
   __pyx_vtable_7sklearn_4tree_5_tree_RegressionCriterion.__pyx_base.eval = (double (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *))__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_eval;
   __pyx_vtable_7sklearn_4tree_5_tree_RegressionCriterion.__pyx_base.init_value = (PyArrayObject *(*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *, int __pyx_skip_dispatch))__pyx_f_7sklearn_4tree_5_tree_19RegressionCriterion_init_value;
   __pyx_type_7sklearn_4tree_5_tree_RegressionCriterion.tp_base = __pyx_ptype_7sklearn_4tree_5_tree_Criterion;
-  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_RegressionCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_RegressionCriterion.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_RegressionCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetAttrString(__pyx_m, "RegressionCriterion", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_RegressionCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 245; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_RegressionCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 348; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_RegressionCriterion.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_RegressionCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 348; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetAttrString(__pyx_m, "RegressionCriterion", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_RegressionCriterion) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 348; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_ptype_7sklearn_4tree_5_tree_RegressionCriterion = &__pyx_type_7sklearn_4tree_5_tree_RegressionCriterion;
   __pyx_vtabptr_7sklearn_4tree_5_tree_MSE = &__pyx_vtable_7sklearn_4tree_5_tree_MSE;
   __pyx_vtable_7sklearn_4tree_5_tree_MSE.__pyx_base = *__pyx_vtabptr_7sklearn_4tree_5_tree_RegressionCriterion;
   __pyx_vtable_7sklearn_4tree_5_tree_MSE.__pyx_base.__pyx_base.eval = (double (*)(struct __pyx_obj_7sklearn_4tree_5_tree_Criterion *))__pyx_f_7sklearn_4tree_5_tree_3MSE_eval;
   __pyx_type_7sklearn_4tree_5_tree_MSE.tp_base = __pyx_ptype_7sklearn_4tree_5_tree_RegressionCriterion;
-  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_MSE) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 394; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_MSE.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_MSE) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 394; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  if (__Pyx_SetAttrString(__pyx_m, "MSE", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_MSE) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 394; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (PyType_Ready(&__pyx_type_7sklearn_4tree_5_tree_MSE) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 576; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetVtable(__pyx_type_7sklearn_4tree_5_tree_MSE.tp_dict, __pyx_vtabptr_7sklearn_4tree_5_tree_MSE) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 576; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  if (__Pyx_SetAttrString(__pyx_m, "MSE", (PyObject *)&__pyx_type_7sklearn_4tree_5_tree_MSE) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 576; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __pyx_ptype_7sklearn_4tree_5_tree_MSE = &__pyx_type_7sklearn_4tree_5_tree_MSE;
   /*--- Type import code ---*/
   __pyx_ptype_5numpy_dtype = __Pyx_ImportType("numpy", "dtype", sizeof(PyArray_Descr), 0); if (unlikely(!__pyx_ptype_5numpy_dtype)) {__pyx_filename = __pyx_f[1]; __pyx_lineno = 154; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
@@ -10657,87 +11961,103 @@ PyMODINIT_FUNC PyInit__tree(void)
   if (PyObject_SetAttr(__pyx_m, __pyx_n_s__DTYPE, __pyx_t_2) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 16; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
 
-  /* "sklearn/tree/_tree.pyx":409
+  /* "sklearn/tree/_tree.pyx":32
+ *     cdef extern double DBL_MAX
  * 
+ * cdef DTYPE_t INFINITY = np.inf             # <<<<<<<<<<<<<<
  * 
- * def _random_sample_mask(int n_total_samples, int n_total_in_bag, random_state):             # <<<<<<<<<<<<<<
- *     """Create a random sample mask where ``n_total_in_bag`` elements are set.
  * 
  */
-  __pyx_t_2 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_1_random_sample_mask, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_2 = __Pyx_GetName(__pyx_m, __pyx_n_s__np); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 32; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
   __Pyx_GOTREF(__pyx_t_2);
-  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___random_sample_mask, __pyx_t_2) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 409; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __pyx_t_1 = PyObject_GetAttr(__pyx_t_2, __pyx_n_s__inf); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 32; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
   __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+  __pyx_t_3 = __pyx_PyFloat_AsFloat(__pyx_t_1); if (unlikely((__pyx_t_3 == (npy_float32)-1) && PyErr_Occurred())) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 32; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
+  __pyx_v_7sklearn_4tree_5_tree_INFINITY = __pyx_t_3;
 
-  /* "sklearn/tree/_tree.pyx":442
+  /* "sklearn/tree/_tree.pyx":604
+ * 
+ * 
+ * def _random_sample_mask(int n_total_samples, int n_total_in_bag, random_state):             # <<<<<<<<<<<<<<
+ *     """Create a random sample mask where ``n_total_in_bag`` elements are set.
+ * 
+ */
+  __pyx_t_1 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_1_random_sample_mask, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___random_sample_mask, __pyx_t_1) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 604; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
+
+  /* "sklearn/tree/_tree.pyx":637
  * 
  * 
  * def _apply_tree(np.ndarray[DTYPE_t, ndim=2] X,             # <<<<<<<<<<<<<<
  *                 np.ndarray[np.int32_t, ndim=2] children,
  *                 np.ndarray[np.int32_t, ndim=1] feature,
  */
-  __pyx_t_2 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_3_apply_tree, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___apply_tree, __pyx_t_2) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 442; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+  __pyx_t_1 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_3_apply_tree, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___apply_tree, __pyx_t_1) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 637; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
 
-  /* "sklearn/tree/_tree.pyx":463
+  /* "sklearn/tree/_tree.pyx":658
  * 
  * 
  * def _predict_tree(np.ndarray[DTYPE_t, ndim=2] X,             # <<<<<<<<<<<<<<
  *                   np.ndarray[np.int32_t, ndim=2] children,
  *                   np.ndarray[np.int32_t, ndim=1] feature,
  */
-  __pyx_t_2 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_5_predict_tree, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___predict_tree, __pyx_t_2) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 463; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+  __pyx_t_1 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_5_predict_tree, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___predict_tree, __pyx_t_1) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 658; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
 
-  /* "sklearn/tree/_tree.pyx":486
+  /* "sklearn/tree/_tree.pyx":685
  * 
  * 
- * def _error_at_leaf(np.ndarray[DTYPE_t, ndim=1, mode="c"] y,             # <<<<<<<<<<<<<<
- *                    np.ndarray sample_mask, Criterion criterion,
- *                    int n_samples):
+ * def _error_at_leaf(np.ndarray[DTYPE_t, ndim=2, mode="c"] y,             # <<<<<<<<<<<<<<
+ *                    np.ndarray sample_mask,
+ *                    Criterion criterion,
  */
-  __pyx_t_2 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_7_error_at_leaf, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___error_at_leaf, __pyx_t_2) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 486; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+  __pyx_t_1 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_7_error_at_leaf, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___error_at_leaf, __pyx_t_1) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 685; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
 
-  /* "sklearn/tree/_tree.pyx":533
+  /* "sklearn/tree/_tree.pyx":736
  * 
  * 
  * def _find_best_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,             # <<<<<<<<<<<<<<
- *                      np.ndarray[DTYPE_t, ndim=1, mode="c"] y,
+ *                      np.ndarray[DTYPE_t, ndim=2, mode="c"] y,
  *                      np.ndarray[np.int32_t, ndim=2, mode="fortran"] X_argsorted,
  */
-  __pyx_t_2 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_9_find_best_split, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___find_best_split, __pyx_t_2) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 533; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+  __pyx_t_1 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_9_find_best_split, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  if (PyObject_SetAttr(__pyx_m, __pyx_n_s___find_best_split, __pyx_t_1) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 736; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
 
-  /* "sklearn/tree/_tree.pyx":678
+  /* "sklearn/tree/_tree.pyx":882
  *     return best_i, best_t, best_error, initial_error
  * 
  * def _find_best_random_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,             # <<<<<<<<<<<<<<
- *                             np.ndarray[DTYPE_t, ndim=1, mode="c"] y,
+ *                             np.ndarray[DTYPE_t, ndim=2, mode="c"] y,
  *                             np.ndarray[np.int32_t, ndim=2, mode="fortran"] X_argsorted,
  */
-  __pyx_t_2 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_11_find_best_random_split, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(__pyx_t_2);
-  if (PyObject_SetAttr(__pyx_m, __pyx_n_s_29, __pyx_t_2) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 678; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(__pyx_t_2); __pyx_t_2 = 0;
+  __pyx_t_1 = PyCFunction_NewEx(&__pyx_mdef_7sklearn_4tree_5_tree_11_find_best_random_split, NULL, __pyx_n_s_16); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(__pyx_t_1);
+  if (PyObject_SetAttr(__pyx_m, __pyx_n_s_29, __pyx_t_1) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 882; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(__pyx_t_1); __pyx_t_1 = 0;
 
   /* "sklearn/tree/_tree.pyx":1
  * # encoding: utf-8             # <<<<<<<<<<<<<<
  * # cython: cdivision=True
  * # cython: boundscheck=False
  */
-  __pyx_t_2 = PyDict_New(); if (unlikely(!__pyx_t_2)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_GOTREF(((PyObject *)__pyx_t_2));
-  if (PyObject_SetAttr(__pyx_m, __pyx_n_s____test__, ((PyObject *)__pyx_t_2)) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
-  __Pyx_DECREF(((PyObject *)__pyx_t_2)); __pyx_t_2 = 0;
+  __pyx_t_1 = PyDict_New(); if (unlikely(!__pyx_t_1)) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_GOTREF(((PyObject *)__pyx_t_1));
+  if (PyObject_SetAttr(__pyx_m, __pyx_n_s____test__, ((PyObject *)__pyx_t_1)) < 0) {__pyx_filename = __pyx_f[0]; __pyx_lineno = 1; __pyx_clineno = __LINE__; goto __pyx_L1_error;}
+  __Pyx_DECREF(((PyObject *)__pyx_t_1)); __pyx_t_1 = 0;
 
   /* "numpy.pxd":974
  *      arr.base = baseptr
@@ -10782,21 +12102,6 @@ static __Pyx_RefNannyAPIStruct *__Pyx_RefNannyImportAPI(const char *modname) {
 }
 #endif /* CYTHON_REFNANNY */
 
-static PyObject *__Pyx_GetName(PyObject *dict, PyObject *name) {
-    PyObject *result;
-    result = PyObject_GetAttr(dict, name);
-    if (!result) {
-        if (dict != __pyx_b) {
-            PyErr_Clear();
-            result = PyObject_GetAttr(__pyx_b, name);
-        }
-        if (!result) {
-            PyErr_SetObject(PyExc_NameError, name);
-        }
-    }
-    return result;
-}
-
 static CYTHON_INLINE int __Pyx_TypeTest(PyObject *obj, PyTypeObject *type) {
     if (unlikely(!type)) {
         PyErr_Format(PyExc_SystemError, "Missing type object");
@@ -10809,6 +12114,31 @@ static CYTHON_INLINE int __Pyx_TypeTest(PyObject *obj, PyTypeObject *type) {
     return 0;
 }
 
+static void __Pyx_RaiseArgtupleInvalid(
+    const char* func_name,
+    int exact,
+    Py_ssize_t num_min,
+    Py_ssize_t num_max,
+    Py_ssize_t num_found)
+{
+    Py_ssize_t num_expected;
+    const char *more_or_less;
+    if (num_found < num_min) {
+        num_expected = num_min;
+        more_or_less = "at least";
+    } else {
+        num_expected = num_max;
+        more_or_less = "at most";
+    }
+    if (exact) {
+        more_or_less = "exactly";
+    }
+    PyErr_Format(PyExc_TypeError,
+                 "%s() takes %s %"PY_FORMAT_SIZE_T"d positional argument%s (%"PY_FORMAT_SIZE_T"d given)",
+                 func_name, more_or_less, num_expected,
+                 (num_expected == 1) ? "" : "s", num_found);
+}
+
 static void __Pyx_RaiseDoubleKeywordsError(
     const char* func_name,
     PyObject* kw_name)
@@ -10899,29 +12229,21 @@ static int __Pyx_ParseOptionalKeywords(
     return -1;
 }
 
-static void __Pyx_RaiseArgtupleInvalid(
-    const char* func_name,
-    int exact,
-    Py_ssize_t num_min,
-    Py_ssize_t num_max,
-    Py_ssize_t num_found)
-{
-    Py_ssize_t num_expected;
-    const char *more_or_less;
-    if (num_found < num_min) {
-        num_expected = num_min;
-        more_or_less = "at least";
-    } else {
-        num_expected = num_max;
-        more_or_less = "at most";
-    }
-    if (exact) {
-        more_or_less = "exactly";
+
+
+static PyObject *__Pyx_GetName(PyObject *dict, PyObject *name) {
+    PyObject *result;
+    result = PyObject_GetAttr(dict, name);
+    if (!result) {
+        if (dict != __pyx_b) {
+            PyErr_Clear();
+            result = PyObject_GetAttr(__pyx_b, name);
+        }
+        if (!result) {
+            PyErr_SetObject(PyExc_NameError, name);
+        }
     }
-    PyErr_Format(PyExc_TypeError,
-                 "%s() takes %s %"PY_FORMAT_SIZE_T"d positional argument%s (%"PY_FORMAT_SIZE_T"d given)",
-                 func_name, more_or_less, num_expected,
-                 (num_expected == 1) ? "" : "s", num_found);
+    return result;
 }
 
 static CYTHON_INLINE int __Pyx_IsLittleEndian(void) {
@@ -11492,40 +12814,6 @@ static CYTHON_INLINE void __Pyx_ErrFetch(PyObject **type, PyObject **value, PyOb
 #endif
 }
 
-static CYTHON_INLINE int __Pyx_CheckKeywordStrings(
-    PyObject *kwdict,
-    const char* function_name,
-    int kw_allowed)
-{
-    PyObject* key = 0;
-    Py_ssize_t pos = 0;
-    while (PyDict_Next(kwdict, &pos, &key, 0)) {
-        #if PY_MAJOR_VERSION < 3
-        if (unlikely(!PyString_CheckExact(key)) && unlikely(!PyString_Check(key)))
-        #else
-        if (unlikely(!PyUnicode_Check(key)))
-        #endif
-            goto invalid_keyword_type;
-    }
-    if ((!kw_allowed) && unlikely(key))
-        goto invalid_keyword;
-    return 1;
-invalid_keyword_type:
-    PyErr_Format(PyExc_TypeError,
-        "%s() keywords must be strings", function_name);
-    return 0;
-invalid_keyword:
-    PyErr_Format(PyExc_TypeError,
-    #if PY_MAJOR_VERSION < 3
-        "%s() got an unexpected keyword argument '%s'",
-        function_name, PyString_AsString(key));
-    #else
-        "%s() got an unexpected keyword argument '%U'",
-        function_name, key);
-    #endif
-    return 0;
-}
-
 static int __Pyx_ArgTypeTest(PyObject *obj, PyTypeObject *type, int none_allowed,
     const char *name, int exact)
 {
@@ -11860,58 +13148,6 @@ static void __Pyx_ReleaseBuffer(Py_buffer *view) {
     return module;
 }
 
-static CYTHON_INLINE npy_int32 __Pyx_PyInt_from_py_npy_int32(PyObject* x) {
-    const npy_int32 neg_one = (npy_int32)-1, const_zero = (npy_int32)0;
-    const int is_unsigned = const_zero < neg_one;
-    if (sizeof(npy_int32) == sizeof(char)) {
-        if (is_unsigned)
-            return (npy_int32)__Pyx_PyInt_AsUnsignedChar(x);
-        else
-            return (npy_int32)__Pyx_PyInt_AsSignedChar(x);
-    } else if (sizeof(npy_int32) == sizeof(short)) {
-        if (is_unsigned)
-            return (npy_int32)__Pyx_PyInt_AsUnsignedShort(x);
-        else
-            return (npy_int32)__Pyx_PyInt_AsSignedShort(x);
-    } else if (sizeof(npy_int32) == sizeof(int)) {
-        if (is_unsigned)
-            return (npy_int32)__Pyx_PyInt_AsUnsignedInt(x);
-        else
-            return (npy_int32)__Pyx_PyInt_AsSignedInt(x);
-    } else if (sizeof(npy_int32) == sizeof(long)) {
-        if (is_unsigned)
-            return (npy_int32)__Pyx_PyInt_AsUnsignedLong(x);
-        else
-            return (npy_int32)__Pyx_PyInt_AsSignedLong(x);
-    } else if (sizeof(npy_int32) == sizeof(PY_LONG_LONG)) {
-        if (is_unsigned)
-            return (npy_int32)__Pyx_PyInt_AsUnsignedLongLong(x);
-        else
-            return (npy_int32)__Pyx_PyInt_AsSignedLongLong(x);
-    }  else {
-        npy_int32 val;
-        PyObject *v = __Pyx_PyNumber_Int(x);
-        #if PY_VERSION_HEX < 0x03000000
-        if (likely(v) && !PyLong_Check(v)) {
-            PyObject *tmp = v;
-            v = PyNumber_Long(tmp);
-            Py_DECREF(tmp);
-        }
-        #endif
-        if (likely(v)) {
-            int one = 1; int is_little = (int)*(unsigned char *)&one;
-            unsigned char *bytes = (unsigned char *)&val;
-            int ret = _PyLong_AsByteArray((PyLongObject *)v,
-                                          bytes, sizeof(val),
-                                          is_little, !is_unsigned);
-            Py_DECREF(v);
-            if (likely(!ret))
-                return val;
-        }
-        return (npy_int32)-1;
-    }
-}
-
 #if CYTHON_CCOMPLEX
   #ifdef __cplusplus
     static CYTHON_INLINE __pyx_t_float_complex __pyx_t_float_complex_from_parts(float x, float y) {
diff --git a/sklearn/tree/_tree.pyx b/sklearn/tree/_tree.pyx
index 55472ddfed261..202ca22bf6dde 100644
--- a/sklearn/tree/_tree.pyx
+++ b/sklearn/tree/_tree.pyx
@@ -17,6 +17,11 @@ DTYPE = np.float32
 ctypedef np.float32_t DTYPE_t
 ctypedef np.int8_t BOOL_t
 
+cdef extern from "stdlib.h":
+    void* malloc(size_t size)
+    void* calloc(size_t nmemb, size_t size)
+    void free(void* ptr)
+
 cdef extern from "math.h":
     cdef extern double log(double x)
     cdef extern double pow(double base, double exponent)
@@ -24,6 +29,8 @@ cdef extern from "math.h":
 cdef extern from "float.h":
     cdef extern double DBL_MAX
 
+cdef DTYPE_t INFINITY = np.inf
+
 
 ################################################################################
 # Classification entropy measures
@@ -40,8 +47,11 @@ cdef extern from "float.h":
 cdef class Criterion:
     """Interface for splitting criteria (regression and classification)"""
 
-    cdef void init(self, DTYPE_t *y, BOOL_t *sample_mask, int n_samples,
-                   int n_total_samples):
+    cdef void init(self, DTYPE_t* y,
+                         int y_stride,
+                         BOOL_t* sample_mask,
+                         int n_samples,
+                         int n_total_samples):
         """Initialise the criterion class for new split point."""
         pass
 
@@ -49,8 +59,12 @@ cdef class Criterion:
         """Reset the criterion for a new feature index."""
         pass
 
-    cdef int update(self, int a, int b, DTYPE_t *y, int *X_argsorted_i,
-                    BOOL_t *sample_mask):
+    cdef int update(self, int a,
+                          int b,
+                          DTYPE_t* y,
+                          int y_stride,
+                          int* X_argsorted_i,
+                          BOOL_t* sample_mask):
         """Update the criteria for each value in interval [a,b) (where a and b
            are indices in `X_argsorted_i`)."""
         pass
@@ -91,92 +105,152 @@ cdef class ClassificationCriterion(Criterion):
     n_right : int
         The number of samples right of splitting point.
     """
-    cdef int n_classes
+    cdef int n_outputs
+    cdef int* n_classes
     cdef int n_samples
+
+    cdef int label_count_stride
     cdef int* label_count_left
     cdef int* label_count_right
     cdef int* label_count_init
+
     cdef int n_left
     cdef int n_right
 
-    # need to store ref to arrays to prevent GC
-    cdef ndarray_label_count_left
-    cdef ndarray_label_count_right
-    cdef ndarray_label_count_init
+    def __init__(self, int n_outputs, object n_classes):
+        cdef int k = 0
+
+        self.n_outputs = n_outputs
+        self.n_classes = <int*> calloc(n_outputs, sizeof(int))
+        cdef int label_count_stride = -1
 
-    def __init__(self, int n_classes):
-        cdef np.ndarray[np.int32_t, ndim=1] ndarray_label_count_left \
-            = np.zeros((n_classes,), dtype=np.int32, order='C')
-        cdef np.ndarray[np.int32_t, ndim=1] ndarray_label_count_right \
-            = np.zeros((n_classes,), dtype=np.int32, order='C')
-        cdef np.ndarray[np.int32_t, ndim=1] ndarray_label_count_init \
-            = np.zeros((n_classes,), dtype=np.int32, order='C')
+        for k from 0 <= k < n_outputs:
+            self.n_classes[k] = n_classes[k]
+
+            if n_classes[k] > label_count_stride:
+                label_count_stride = n_classes[k]
+
+        self.label_count_stride = label_count_stride
+        self.label_count_left = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+        self.label_count_right = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
+        self.label_count_init = <int*> calloc(n_outputs * label_count_stride, sizeof(int))
 
-        self.n_classes = n_classes
         self.n_samples = 0
         self.n_left = 0
         self.n_right = 0
-        self.label_count_left = <int*>ndarray_label_count_left.data
-        self.label_count_right = <int*>ndarray_label_count_right.data
-        self.label_count_init = <int*>ndarray_label_count_init.data
-        self.ndarray_label_count_left = ndarray_label_count_left
-        self.ndarray_label_count_right = ndarray_label_count_right
-        self.ndarray_label_count_init = ndarray_label_count_init
-
-    cdef void init(self, DTYPE_t *y, BOOL_t *sample_mask, int n_samples,
-                   int n_total_samples):
+
+    def __del__(self):
+        free(self.n_classes)
+        free(self.label_count_left)
+        free(self.label_count_right)
+        free(self.label_count_init)
+
+    cdef void init(self, DTYPE_t* y,
+                         int y_stride,
+                         BOOL_t *sample_mask,
+                         int n_samples,
+                         int n_total_samples):
         """Initialise the criterion class."""
+        cdef int n_outputs = self.n_outputs
+        cdef int* n_classes = self.n_classes
+        cdef int label_count_stride = self.label_count_stride
+        cdef int* label_count_init = self.label_count_init
+
+        cdef int k = 0
         cdef int c = 0
         cdef int j = 0
 
         self.n_samples = n_samples
 
-        for c from 0 <= c < self.n_classes:
-            self.label_count_init[c] = 0
+        for k from 0 <= k < n_outputs:
+            for c from 0 <= c < n_classes[k]:
+                label_count_init[k * label_count_stride + c] = 0
 
         for j from 0 <= j < n_total_samples:
             if sample_mask[j] == 0:
                 continue
-            c = <int>(y[j])
-            self.label_count_init[c] += 1
+
+            for k from 0 <= k < n_outputs:
+                c = <int>y[j * y_stride + k]
+                label_count_init[k * label_count_stride + c] += 1
 
         self.reset()
 
     cdef void reset(self):
         """Reset label_counts by setting `label_count_left to zero
         and copying the init array into the right."""
+        cdef int n_outputs = self.n_outputs
+        cdef int* n_classes = self.n_classes
+        cdef int label_count_stride = self.label_count_stride
+        cdef int* label_count_init = self.label_count_init
+        cdef int* label_count_left = self.label_count_left
+        cdef int* label_count_right = self.label_count_right
+
+        cdef int k = 0
         cdef int c = 0
         self.n_left = 0
         self.n_right = self.n_samples
 
-        for c from 0 <= c < self.n_classes:
-            self.label_count_left[c] = 0
-            self.label_count_right[c] = self.label_count_init[c]
-
-    cdef int update(self, int a, int b, DTYPE_t *y, int *X_argsorted_i,
-                    BOOL_t *sample_mask):
+        for k from 0 <= k < n_outputs:
+            for c from 0 <= c < n_classes[k]:
+                label_count_left[k * label_count_stride + c] = 0
+                label_count_right[k * label_count_stride + c] = label_count_init[k * label_count_stride + c]
+
+    cdef int update(self, int a,
+                          int b,
+                          DTYPE_t* y,
+                          int y_stride,
+                          int* X_argsorted_i,
+                          BOOL_t* sample_mask):
         """Update the criteria for each value in interval [a,b) (where a and b
            are indices in `X_argsorted_i`)."""
-        cdef int c
+        cdef int n_outputs = self.n_outputs
+        cdef int label_count_stride = self.label_count_stride
+        cdef int* label_count_left = self.label_count_left
+        cdef int* label_count_right = self.label_count_right
+        cdef int n_left = self.n_left
+        cdef int n_right = self.n_right
+
+        cdef int idx, k, c, s
+
         # post condition: all samples from [0:b) are on the left side
         for idx from a <= idx < b:
             s = X_argsorted_i[idx]
+
             if sample_mask[s] == 0:
                 continue
-            c = <int>(y[s])
-            self.label_count_right[c] -= 1
-            self.label_count_left[c] += 1
-            self.n_right -= 1
-            self.n_left += 1
 
-        return self.n_left
+            for k from 0 <= k < n_outputs:
+                c = <int>y[s * y_stride + k]
+                label_count_right[k * label_count_stride + c] -= 1
+                label_count_left[k * label_count_stride + c] += 1
+
+            n_left += 1
+            n_right -=1
+
+        self.n_left = n_left
+        self.n_right = n_right
+
+        return n_left
 
     cdef double eval(self):
         pass
 
     cpdef np.ndarray init_value(self):
-        return self.ndarray_label_count_init
+        cdef int n_outputs = self.n_outputs
+        cdef int* n_classes = self.n_classes
+        cdef int label_count_stride = self.label_count_stride
+        cdef int* label_count_init = self.label_count_init
 
+        cdef np.ndarray[DTYPE_t, ndim=2] value = np.zeros((n_outputs, label_count_stride), dtype=DTYPE)
+
+        cdef int k, c
+
+        for k from 0 <= k < n_outputs:
+            for c from 0 <= c < n_classes[k]:
+                value[k, c] = <DTYPE_t> (label_count_init[k * label_count_stride + c])
+
+        return value
 
 cdef class Gini(ClassificationCriterion):
     """Gini Index splitting criteria.
@@ -187,31 +261,46 @@ cdef class Gini(ClassificationCriterion):
 
     cdef double eval(self):
         """Returns Gini index of left branch + Gini index of right branch. """
+        cdef int n_samples = self.n_samples
+        cdef int n_outputs = self.n_outputs
+        cdef int* n_classes = self.n_classes
+        cdef int label_count_stride = self.label_count_stride
+        cdef int* label_count_left = self.label_count_left
+        cdef int* label_count_right = self.label_count_right
         cdef double n_left = <double> self.n_left
         cdef double n_right = <double> self.n_right
-        cdef double H_left = n_left * n_left
-        cdef double H_right = n_right * n_right
-        cdef int k, count_left, count_right
 
-        for k from 0 <= k < self.n_classes:
-            count_left = self.label_count_left[k]
-            if count_left > 0:
-                H_left -= (count_left * count_left)
-            count_right = self.label_count_right[k]
-            if count_right > 0:
-                H_right -= (count_right * count_right)
+        cdef double total = 0.0
+        cdef double H_left
+        cdef double H_right
+        cdef int k, c, count_left, count_right
+
+        for k from 0 <= k < n_outputs:
+            H_left = n_left * n_left
+            H_right = n_right * n_right
+
+            for c from 0 <= c < n_classes[k]:
+                count_left = label_count_left[k * label_count_stride + c]
+                if count_left > 0:
+                    H_left -= (count_left * count_left)
+
+                count_right = label_count_right[k * label_count_stride + c]
+                if count_right > 0:
+                    H_right -= (count_right * count_right)
+
+            if n_left == 0:
+                H_left = 0
+            else:
+                H_left /= n_left
 
-        if n_left == 0:
-            H_left = 0
-        else:
-            H_left /= n_left
+            if n_right == 0:
+                H_right = 0
+            else:
+                H_right /= n_right
 
-        if n_right == 0:
-            H_right = 0
-        else:
-            H_right /= n_right
+            total += (H_left + H_right)
 
-        return (H_left + H_right) / self.n_samples
+        return total / (n_samples * n_outputs)
 
 
 cdef class Entropy(ClassificationCriterion):
@@ -222,24 +311,38 @@ cdef class Entropy(ClassificationCriterion):
 
     cdef double eval(self):
         """Returns Entropy of left branch + Entropy index of right branch. """
-        cdef double H_left = 0.0
-        cdef double H_right = 0.0
-        cdef int k
-        cdef double e1, e2
+        cdef int n_samples = self.n_samples
+        cdef int n_outputs = self.n_outputs
+        cdef int* n_classes = self.n_classes
+        cdef int label_count_stride = self.label_count_stride
+        cdef int* label_count_left = self.label_count_left
+        cdef int* label_count_right = self.label_count_right
         cdef double n_left = <double> self.n_left
         cdef double n_right = <double> self.n_right
 
-        for k from 0 <= k < self.n_classes:
-            if self.label_count_left[k] > 0:
-                H_left -= ((self.label_count_left[k] / n_left)
-                           * log(self.label_count_left[k] / n_left))
-            if self.label_count_right[k] > 0:
-                H_right -= ((self.label_count_right[k] / n_right)
-                            * log(self.label_count_right[k] / n_right))
+        cdef double total = 0.0
+        cdef double H_left
+        cdef double H_right
+        cdef int k, c
+        cdef double e1, e2
+
+        for k from 0 <= k < n_outputs:
+            H_left = 0.0
+            H_right = 0.0
+
+            for c from 0 <= c < n_classes[k]:
+                if label_count_left[k * label_count_stride + c] > 0:
+                    H_left -= ((label_count_left[k * label_count_stride + c] / n_left) * log(label_count_left[k * label_count_stride + c] / n_left))
+
+                if self.label_count_right[k * label_count_stride + c] > 0:
+                    H_right -= ((label_count_right[k * label_count_stride + c] / n_right) * log(label_count_right[k * label_count_stride + c] / n_right))
+
+            e1 = (n_left / n_samples) * H_left
+            e2 = (n_right / n_samples) * H_right
 
-        e1 = (n_left / self.n_samples) * H_left
-        e2 = (n_right / self.n_samples) * H_right
-        return e1 + e2
+            total += e1 + e2
+
+        return total / n_outputs
 
 
 cdef class RegressionCriterion(Criterion):
@@ -281,57 +384,95 @@ cdef class RegressionCriterion(Criterion):
         number of samples right of split point.
     """
 
+    cdef int n_outputs
     cdef int n_samples
+
+    cdef double* mean_left
+    cdef double* mean_right
+    cdef double* mean_init
+    cdef double* sq_sum_left
+    cdef double* sq_sum_right
+    cdef double* sq_sum_init
+    cdef double* var_left
+    cdef double* var_right
+
     cdef int n_right
     cdef int n_left
 
-    cdef double mean_left
-    cdef double mean_right
-    cdef double mean_init
+    def __init__(self, int n_outputs):
+        cdef int k = 0
 
-    cdef double sq_sum_right
-    cdef double sq_sum_left
-    cdef double sq_sum_init
+        self.n_outputs = n_outputs
 
-    cdef double var_left
-    cdef double var_right
-
-    def __init__(self):
         self.n_samples = 0
         self.n_left = 0
         self.n_right = 0
-        self.mean_left = 0.0
-        self.mean_right = 0.0
-        self.mean_init = 0.0
-        self.sq_sum_right = 0.0
-        self.sq_sum_left = 0.0
-        self.sq_sum_init = 0.0
-        self.var_left = 0.0
-        self.var_right = 0.0
-
-    cdef void init(self, DTYPE_t *y, BOOL_t *sample_mask, int n_samples,
-                   int n_total_samples):
+
+        self.mean_left = <double*> calloc(n_outputs, sizeof(double))
+        self.mean_right = <double*> calloc(n_outputs, sizeof(double))
+        self.mean_init = <double*> calloc(n_outputs, sizeof(double))
+        self.sq_sum_left = <double*> calloc(n_outputs, sizeof(double))
+        self.sq_sum_right = <double*> calloc(n_outputs, sizeof(double))
+        self.sq_sum_init = <double*> calloc(n_outputs, sizeof(double))
+        self.var_left = <double*> calloc(n_outputs, sizeof(double))
+        self.var_right = <double*> calloc(n_outputs, sizeof(double))
+
+    def __del__(self):
+        free(self.mean_left)
+        free(self.mean_right)
+        free(self.mean_init)
+        free(self.sq_sum_left)
+        free(self.sq_sum_right)
+        free(self.sq_sum_init)
+        free(self.var_left)
+        free(self.var_right)
+
+    cdef void init(self, DTYPE_t* y,
+                         int y_stride,
+                         BOOL_t* sample_mask,
+                         int n_samples,
+                         int n_total_samples):
         """Initialise the criterion class; assume all samples
            are in the right branch and store the mean and squared
            sum in `self.mean_init` and `self.sq_sum_init`. """
-        self.mean_left = 0.0
-        self.mean_right = 0.0
-        self.mean_init = 0.0
-        self.sq_sum_right = 0.0
-        self.sq_sum_left = 0.0
-        self.sq_sum_init = 0.0
-        self.var_left = 0.0
-        self.var_right = 0.0
+        cdef double* mean_left = self.mean_left
+        cdef double* mean_right = self.mean_right
+        cdef double* mean_init = self.mean_init
+        cdef double* sq_sum_left = self.sq_sum_left
+        cdef double* sq_sum_right = self.sq_sum_right
+        cdef double* sq_sum_init = self.sq_sum_init
+        cdef double* var_left = self.var_left
+        cdef double* var_right = self.var_right
+        cdef int n_outputs = self.n_outputs
+
+        cdef int k = 0
+
+        for k from 0 <= k < n_outputs:
+            mean_left[k] = 0.0
+            mean_right[k] = 0.0
+            mean_init[k] = 0.0
+            sq_sum_right[k] = 0.0
+            sq_sum_left[k] = 0.0
+            sq_sum_init[k] = 0.0
+            var_left[k] = 0.0
+            var_right[k] = 0.0
+
         self.n_samples = n_samples
 
         cdef int j = 0
+        cdef DTYPE_t y_jk = 0.0
+
         for j from 0 <= j < n_total_samples:
             if sample_mask[j] == 0:
                 continue
-            self.sq_sum_init += (y[j] * y[j])
-            self.mean_init += y[j]
 
-        self.mean_init = self.mean_init / self.n_samples
+            for k from 0 <= k < n_outputs:
+                y_jk = y[j * y_stride + k]
+                sq_sum_init[k] += y_jk * y_jk
+                mean_init[k] += y_jk
+
+        for k from 0 <= k < n_outputs:
+            mean_init[k] /= n_samples
 
         self.reset()
 
@@ -342,53 +483,94 @@ cdef class RegressionCriterion(Criterion):
         whole dataset into the auxiliary variables of the
         right branch.
         """
+        cdef double* mean_left = self.mean_left
+        cdef double* mean_right = self.mean_right
+        cdef double* mean_init = self.mean_init
+        cdef double* sq_sum_left = self.sq_sum_left
+        cdef double* sq_sum_right = self.sq_sum_right
+        cdef double* sq_sum_init = self.sq_sum_init
+        cdef double* var_left = self.var_left
+        cdef double* var_right = self.var_right
+
+        cdef int n_samples = self.n_samples
+        cdef int n_outputs = self.n_outputs
+
+        cdef int k = 0
+
         self.n_right = self.n_samples
         self.n_left = 0
-        self.mean_right = self.mean_init
-        self.mean_left = 0.0
-        self.sq_sum_right = self.sq_sum_init
-        self.sq_sum_left = 0.0
-        self.var_left = 0.0
-        self.var_right = self.sq_sum_right - \
-            self.n_samples * (self.mean_right * self.mean_right)
-
-    cdef int update(self, int a, int b, DTYPE_t *y, int *X_argsorted_i,
-                    BOOL_t *sample_mask):
+
+        for k from 0 <= k < n_outputs:
+            mean_right[k] = mean_init[k]
+            mean_left[k] = 0.0
+            sq_sum_right[k] = sq_sum_init[k]
+            sq_sum_left[k] = 0.0
+            var_left[k] = 0.0
+            var_right[k] = sq_sum_right[k] - n_samples * (mean_right[k] * mean_right[k])
+
+    cdef int update(self, int a,
+                          int b,
+                          DTYPE_t* y,
+                          int y_stride,
+                          int* X_argsorted_i,
+                          BOOL_t* sample_mask):
         """Update the criteria for each value in interval [a,b) (where a and b
            are indices in `X_argsorted_i`)."""
+        cdef double* mean_left = self.mean_left
+        cdef double* mean_right = self.mean_right
+        cdef double* sq_sum_left = self.sq_sum_left
+        cdef double* sq_sum_right = self.sq_sum_right
+        cdef double* var_left = self.var_left
+        cdef double* var_right = self.var_right
+
+        cdef int n_samples = self.n_samples
+        cdef int n_outputs = self.n_outputs
+        cdef int n_left = self.n_left
+        cdef int n_right = self.n_right
+
         cdef double y_idx = 0.0
-        cdef int idx, j
+        cdef int idx, j, k
+
         # post condition: all samples from [0:b) are on the left side
         for idx from a <= idx < b:
             j = X_argsorted_i[idx]
+
             if sample_mask[j] == 0:
                 continue
-            y_idx = y[j]
-            self.sq_sum_left = self.sq_sum_left + (y_idx * y_idx)
-            self.sq_sum_right = self.sq_sum_right - (y_idx * y_idx)
 
-            self.mean_left = (self.n_left * self.mean_left + y_idx) / \
-                <double>(self.n_left + 1)
-            self.mean_right = ((self.n_samples - self.n_left) * \
-                self.mean_right - y_idx) / \
-                <double>(self.n_samples - self.n_left - 1)
+            for k from 0 <= k < n_outputs:
+                y_idx = y[j * y_stride + k]
+                sq_sum_left[k] += (y_idx * y_idx)
+                sq_sum_right[k] -= (y_idx * y_idx)
 
-            self.n_right -= 1
-            self.n_left += 1
+                mean_left[k] = (n_left * mean_left[k] + y_idx) / <double>(n_left + 1)
+                mean_right[k] = ((n_samples - n_left) * mean_right[k] - y_idx) / <double>(n_samples - n_left - 1)
 
-            self.var_left = self.sq_sum_left - \
-                self.n_left * (self.mean_left * self.mean_left)
-            self.var_right = self.sq_sum_right - \
-                self.n_right * (self.mean_right * self.mean_right)
+            n_left += 1
+            self.n_left = n_left
+            n_right -= 1
+            self.n_right = n_right
 
-        return self.n_left
+            for k from 0 <= k < n_outputs:
+                var_left[k] = sq_sum_left[k] - n_left * (mean_left[k] * mean_left[k])
+                var_right[k] = sq_sum_right[k] - n_right * (mean_right[k] * mean_right[k])
+
+        return n_left
 
     cdef double eval(self):
         pass
 
     cpdef np.ndarray init_value(self):
-        ## TODO is calling np.asarray a performance issue?
-        return np.asarray(self.mean_init)
+        cdef int n_outputs = self.n_outputs
+        cdef double* mean_init = self.mean_init
+
+        cdef np.ndarray[DTYPE_t, ndim=2] value = np.zeros((n_outputs, 1), dtype=DTYPE)
+        cdef int k
+
+        for k from 0 <= k < n_outputs:
+            value[k, 0] = <DTYPE_t> (mean_init[k])
+
+        return value
 
 
 cdef class MSE(RegressionCriterion):
@@ -398,7 +580,20 @@ cdef class MSE(RegressionCriterion):
     """
 
     cdef double eval(self):
-        return self.var_left + self.var_right
+        cdef double* var_left = self.var_left
+        cdef double* var_right = self.var_right
+
+        cdef int n_outputs = self.n_outputs
+
+        cdef int k
+        cdef double total = 0.0
+
+        for k from 0 <= k < n_outputs:
+            total += var_left[k]
+            total += var_right[k]
+
+        return total / n_outputs
+
 
 
 ################################################################################
@@ -431,7 +626,7 @@ def _random_sample_mask(int n_total_samples, int n_total_in_bag, random_state):
 
     cdef int n_bagged = 0
     cdef int i = 0
-    for i in range(n_total_samples):
+    for i from 0 <= i < n_total_samples:
         if rand[i] * (n_total_samples - i) < (n_total_in_bag - n_bagged):
             sample_mask[i] = 1
             n_bagged += 1
@@ -449,7 +644,7 @@ def _apply_tree(np.ndarray[DTYPE_t, ndim=2] X,
     cdef int i = 0
     cdef int n = X.shape[0]
     cdef int node_id = 0
-    for i in xrange(n):
+    for i from 0 <= i < n:
         node_id = 0
         # While node_id not a leaf
         while children[node_id, 0] != -1 and children[node_id, 1] != -1:
@@ -464,14 +659,16 @@ def _predict_tree(np.ndarray[DTYPE_t, ndim=2] X,
                   np.ndarray[np.int32_t, ndim=2] children,
                   np.ndarray[np.int32_t, ndim=1] feature,
                   np.ndarray[np.float64_t, ndim=1] threshold,
-                  np.ndarray[np.float64_t, ndim=2] values,
-                  np.ndarray[np.float64_t, ndim=2] pred):
+                  np.ndarray[np.float64_t, ndim=3] values,
+                  np.ndarray[np.float64_t, ndim=3] pred):
     """Finds the terminal region (=leaf node) values for each sample. """
-    cdef int i = 0
+    cdef int i, k, c
     cdef int n = X.shape[0]
     cdef int node_id = 0
-    cdef int K = values.shape[1]
-    for i in xrange(n):
+    cdef int n_outputs = values.shape[1]
+    cdef int n_classes = values.shape[2]
+
+    for i from 0 <= i < n:
         node_id = 0
         # While node_id not a leaf
         while children[node_id, 0] != -1 and children[node_id, 1] != -1:
@@ -479,19 +676,25 @@ def _predict_tree(np.ndarray[DTYPE_t, ndim=2] X,
                 node_id = children[node_id, 0]
             else:
                 node_id = children[node_id, 1]
-        for k in xrange(K):
-            pred[i, k] = values[node_id, k]
+
+        for k from 0 <= k < n_outputs:
+            for c from 0 <= c < n_classes:
+                pred[i, k, c] = values[node_id, k, c]
 
 
-def _error_at_leaf(np.ndarray[DTYPE_t, ndim=1, mode="c"] y,
-                   np.ndarray sample_mask, Criterion criterion,
+def _error_at_leaf(np.ndarray[DTYPE_t, ndim=2, mode="c"] y,
+                   np.ndarray sample_mask,
+                   Criterion criterion,
                    int n_samples):
     """Compute criterion error at leaf with terminal region defined
     by `sample_mask`. """
+    cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+    cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
     cdef int n_total_samples = y.shape[0]
-    cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
     cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
-    criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
+
+    criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)
+
     return criterion.eval()
 
 
@@ -531,7 +734,7 @@ cdef int smallest_sample_larger_than(int sample_idx, DTYPE_t *X_i,
 
 
 def _find_best_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
-                     np.ndarray[DTYPE_t, ndim=1, mode="c"] y,
+                     np.ndarray[DTYPE_t, ndim=2, mode="c"] y,
                      np.ndarray[np.int32_t, ndim=2, mode="fortran"] X_argsorted,
                      np.ndarray sample_mask,
                      int n_samples,
@@ -595,15 +798,16 @@ def _find_best_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
     cdef np.int32_t feature_idx = -1
     cdef int n_left = 0
     cdef DTYPE_t t, initial_error, error
-    cdef DTYPE_t best_error = np.inf, best_t = np.inf
-    cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
-    cdef DTYPE_t *X_i = NULL
-    cdef int *X_argsorted_i = NULL
-    cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
-    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] features = None
+    cdef DTYPE_t best_error = INFINITY, best_t = INFINITY
+    cdef DTYPE_t* X_i = NULL
+    cdef int* X_argsorted_i = NULL
+    cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+    cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data
+    cdef np.ndarray[np.int32_t, ndim=1, mode="c"] features = None
 
     # Compute the column strides (increment in pointer elements to get
     # from column i to i + 1) for `X` and `X_argsorted`
+    cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
     cdef int X_elem_stride = X.strides[0]
     cdef int X_col_stride = X.strides[1]
     cdef int X_stride = X_col_stride / X_elem_stride
@@ -612,8 +816,8 @@ def _find_best_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
     cdef int X_argsorted_stride = X_argsorted_col_stride / X_argsorted_elem_stride
 
     # Compute the initial criterion value in the node
-    X_argsorted_i = <int *>X_argsorted.data
-    criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
+    X_argsorted_i = <int*>X_argsorted.data
+    criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)
     initial_error = criterion.eval()
 
     if initial_error == 0:  # break early if the node is pure
@@ -629,11 +833,11 @@ def _find_best_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
         features = random_state.permutation(features)[:max_features]
 
     # Look for the best split
-    for feature_idx in range(max_features):
+    for feature_idx from 0 <= feature_idx < max_features:
         i = features[feature_idx]
         # Get i-th col of X and X_sorted
-        X_i = (<DTYPE_t *>X.data) + X_stride * i
-        X_argsorted_i = (<int *>X_argsorted.data) + X_argsorted_stride * i
+        X_i = (<DTYPE_t*>X.data) + X_stride * i
+        X_argsorted_i = (<int*>X_argsorted.data) + X_argsorted_stride * i
 
         # Reset the criterion for this feature
         criterion.reset()
@@ -652,7 +856,7 @@ def _find_best_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
                 break
 
             # Better split than the best so far?
-            n_left = criterion.update(a, b, y_ptr, X_argsorted_i, sample_mask_ptr)
+            n_left = criterion.update(a, b, y_ptr, y_stride, X_argsorted_i, sample_mask_ptr)
 
             # Only consider splits that respect min_leaf
             if n_left < min_leaf or (n_samples - n_left) < min_leaf:
@@ -676,7 +880,7 @@ def _find_best_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
     return best_i, best_t, best_error, initial_error
 
 def _find_best_random_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
-                            np.ndarray[DTYPE_t, ndim=1, mode="c"] y,
+                            np.ndarray[DTYPE_t, ndim=2, mode="c"] y,
                             np.ndarray[np.int32_t, ndim=2, mode="fortran"] X_argsorted,
                             np.ndarray sample_mask,
                             int n_samples,
@@ -739,15 +943,16 @@ def _find_best_random_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
     cdef int i, a, b, c, n_left, best_i = -1
     cdef np.int32_t feature_idx = -1
     cdef DTYPE_t t, initial_error, error
-    cdef DTYPE_t best_error = np.inf, best_t = np.inf
-    cdef DTYPE_t *y_ptr = <DTYPE_t *>y.data
-    cdef DTYPE_t *X_i = NULL
-    cdef int *X_argsorted_i = NULL
-    cdef BOOL_t *sample_mask_ptr = <BOOL_t *>sample_mask.data
-    cdef np.ndarray[np.int32_t, ndim=1, mode='c'] features = None
+    cdef DTYPE_t best_error = INFINITY, best_t = INFINITY
+    cdef DTYPE_t* X_i = NULL
+    cdef int* X_argsorted_i = NULL
+    cdef DTYPE_t* y_ptr = <DTYPE_t*>y.data
+    cdef BOOL_t* sample_mask_ptr = <BOOL_t*>sample_mask.data
+    cdef np.ndarray[np.int32_t, ndim=1, mode="c"] features = None
 
     # Compute the column strides (increment in pointer elements to get
     # from column i to i + 1) for `X` and `X_argsorted`
+    cdef int y_stride = <int>y.strides[0] / <int>y.strides[1]
     cdef int X_elem_stride = X.strides[0]
     cdef int X_col_stride = X.strides[1]
     cdef int X_stride = X_col_stride / X_elem_stride
@@ -756,8 +961,8 @@ def _find_best_random_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
     cdef int X_argsorted_stride = X_argsorted_col_stride / X_argsorted_elem_stride
 
     # Compute the initial criterion value
-    X_argsorted_i = <int *>X_argsorted.data
-    criterion.init(y_ptr, sample_mask_ptr, n_samples, n_total_samples)
+    X_argsorted_i = <int*>X_argsorted.data
+    criterion.init(y_ptr, y_stride, sample_mask_ptr, n_samples, n_total_samples)
     initial_error = criterion.eval()
 
     if initial_error == 0:  # break early if the node is pure
@@ -773,11 +978,11 @@ def _find_best_random_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
         features = random_state.permutation(features)[:max_features]
 
     # Look for the best random split
-    for feature_idx in range(max_features):
+    for feature_idx from 0 <= feature_idx < max_features:
         i = features[feature_idx]
         # Get i-th col of X and X_sorted
-        X_i = (<DTYPE_t *>X.data) + X_stride * i
-        X_argsorted_i = (<int *>X_argsorted.data) + X_argsorted_stride * i
+        X_i = (<DTYPE_t*>X.data) + X_stride * i
+        X_argsorted_i = (<int*>X_argsorted.data) + X_argsorted_stride * i
 
         # Reset the criterion for this feature
         criterion.reset()
@@ -811,7 +1016,7 @@ def _find_best_random_split(np.ndarray[DTYPE_t, ndim=2, mode="fortran"] X,
             c += 1
 
         # Better than the best so far?
-        n_left = criterion.update(0, c, y_ptr, X_argsorted_i, sample_mask_ptr)
+        n_left = criterion.update(0, c, y_ptr, y_stride, X_argsorted_i, sample_mask_ptr)
         error = criterion.eval()
 
         if n_left < min_leaf or (n_samples - n_left) < min_leaf:
diff --git a/sklearn/tree/tests/test_tree.py b/sklearn/tree/tests/test_tree.py
index 14fe00fb3ee39..8092b4b342127 100644
--- a/sklearn/tree/tests/test_tree.py
+++ b/sklearn/tree/tests/test_tree.py
@@ -348,6 +348,61 @@ def test_pickle():
             " after pickling (regression) "
 
 
+def test_multioutput():
+    """Check estimators on multi-output problems."""
+
+    X = [[-2, -1],
+         [-1, -1],
+         [-1, -2],
+         [1, 1],
+         [1, 2],
+         [2, 1],
+         [-2, 1],
+         [-1, 1],
+         [-1, 2],
+         [2, -1],
+         [1, -1],
+         [1, -2]]
+
+    y = [[-1, 0],
+         [-1, 0],
+         [-1, 0],
+         [1, 1],
+         [1, 1],
+         [1, 1],
+         [-1, 2],
+         [-1, 2],
+         [-1, 2],
+         [1, 3],
+         [1, 3],
+         [1, 3]]
+
+    T = [[-1, -1], [1, 1], [-1, 1], [1, -1]]
+    y_true = [[-1, 0], [1, 1], [-1, 2], [1, 3]]
+
+    # toy classification problem
+    clf = tree.DecisionTreeClassifier()
+    y_hat = clf.fit(X, y).predict(T)
+    assert_array_equal(y_hat, y_true)
+    assert_equal(y_hat.shape, (4, 2))
+
+    proba = clf.predict_proba(T)
+    assert_equal(len(proba), 2)
+    assert_equal(proba[0].shape, (4, 2))
+    assert_equal(proba[1].shape, (4, 4))
+
+    log_proba = clf.predict_log_proba(T)
+    assert_equal(len(log_proba), 2)
+    assert_equal(log_proba[0].shape, (4, 2))
+    assert_equal(log_proba[1].shape, (4, 4))
+
+    # toy regression problem
+    clf = tree.DecisionTreeRegressor()
+    y_hat = clf.fit(X, y).predict(T)
+    assert_almost_equal(y_hat, y_true)
+    assert_equal(y_hat.shape, (4, 2))
+
+
 if __name__ == "__main__":
     import nose
     nose.runmodule()
diff --git a/sklearn/tree/tree.py b/sklearn/tree/tree.py
index 675c4d309ae0d..310b6f6ff68f4 100644
--- a/sklearn/tree/tree.py
+++ b/sklearn/tree/tree.py
@@ -1,6 +1,6 @@
 """
 This module gathers tree-based methods, including decision, regression and
-randomized trees.
+randomized trees. Single and multi-output problems are both handled.
 """
 
 # Code is originally adapted from MILK: Machine Learning Toolkit
@@ -20,6 +20,8 @@
 
 from . import _tree
 
+import sys
+
 __all__ = ["DecisionTreeClassifier",
            "DecisionTreeRegressor",
            "ExtraTreeClassifier",
@@ -82,15 +84,23 @@ def node_to_str(tree, node_id):
             feature = feature_names[tree.feature[node_id]]
         else:
             feature = "X[%s]" % tree.feature[node_id]
+
+        value = tree.value[node_id]
+        if tree.n_outputs == 1:
+            value = value[0, :]
+
         if tree.children[node_id, 0] == Tree.LEAF:
             return "error = %.4f\\nsamples = %s\\nvalue = %s" \
-                   % (tree.init_error[node_id], tree.n_samples[node_id],
-                      tree.value[node_id])
-
-        return "%s <= %.4f\\nerror = %s\\nsamples = %s\\nvalue = %s" \
-               % (feature, tree.threshold[node_id],
-                  tree.init_error[node_id], tree.n_samples[node_id],
-                  tree.value[node_id])
+                   % (tree.init_error[node_id],
+                      tree.n_samples[node_id],
+                      value)
+        else:
+            return "%s <= %.4f\\nerror = %s\\nsamples = %s\\nvalue = %s" \
+                   % (feature,
+                      tree.threshold[node_id],
+                      tree.init_error[node_id],
+                      tree.n_samples[node_id],
+                      value)
 
     def recurse(tree, node_id, parent=None):
         if node_id == Tree.LEAF:
@@ -145,9 +155,9 @@ class Tree(object):
         The feature to split on (only for internal nodes).
 
     threshold : np.ndarray of float64
-        The threshold of each node (only for leaves).
+        The threshold of each node (only for internal nodes).
 
-    value : np.ndarray of float64, shape=(capacity, n_classes)
+    value : np.ndarray of float64, shape=(capacity, n_outputs, n_classes)
         Contains the constant prediction value of each node.
 
     best_error : np.ndarray of float64
@@ -165,9 +175,10 @@ class Tree(object):
     LEAF = -1
     UNDEFINED = -2
 
-    def __init__(self, n_classes, n_features, capacity=3):
+    def __init__(self, n_classes, n_features, n_outputs, capacity=3):
         self.n_classes = n_classes
         self.n_features = n_features
+        self.n_outputs = n_outputs
 
         self.node_count = 0
 
@@ -178,7 +189,8 @@ def __init__(self, n_classes, n_features, capacity=3):
         self.feature.fill(Tree.UNDEFINED)
 
         self.threshold = np.empty((capacity,), dtype=np.float64)
-        self.value = np.empty((capacity, n_classes), dtype=np.float64)
+        self.value = np.empty((capacity, n_outputs, np.max(n_classes)),
+                              dtype=np.float64)
 
         self.best_error = np.empty((capacity,), dtype=np.float32)
         self.init_error = np.empty((capacity,), dtype=np.float32)
@@ -195,7 +207,8 @@ def _resize(self, capacity=None):
         self.children.resize((capacity, 2), refcheck=False)
         self.feature.resize((capacity,), refcheck=False)
         self.threshold.resize((capacity,), refcheck=False)
-        self.value.resize((capacity, self.value.shape[1]), refcheck=False)
+        self.value.resize((capacity, self.value.shape[1], self.value.shape[2]),
+                          refcheck=False)
         self.best_error.resize((capacity,), refcheck=False)
         self.init_error.resize((capacity,), refcheck=False)
         self.n_samples.resize((capacity,), refcheck=False)
@@ -218,6 +231,7 @@ def _add_split_node(self, parent, is_left_child, feature, threshold,
         self.init_error[node_id] = init_error
         self.best_error[node_id] = best_error
         self.n_samples[node_id] = n_samples
+
         self.value[node_id] = value
 
         # set as left or right child of parent
@@ -228,6 +242,7 @@ def _add_split_node(self, parent, is_left_child, feature, threshold,
                 self.children[parent, 1] = node_id
 
         self.node_count += 1
+
         return node_id
 
     def _add_leaf(self, parent, is_left_child, value, error, n_samples):
@@ -248,8 +263,8 @@ def _add_leaf(self, parent, is_left_child, value, error, n_samples):
             self.children[parent, 1] = node_id
 
         self.children[node_id, :] = Tree.LEAF
-
         self.node_count += 1
+
         return node_id
 
     def build(self, X, y, criterion, max_depth, min_samples_split,
@@ -315,10 +330,10 @@ def recursive_partition(X, X_argsorted, y, sample_mask, depth,
         # Setup auxiliary data structures and check input before
         # recursive partitioning
         if X.dtype != DTYPE or not np.isfortran(X):
-            X = np.asanyarray(X, dtype=DTYPE, order="F")
+            X = np.asarray(X, dtype=DTYPE, order="F")
 
         if y.dtype != DTYPE or not y.flags.contiguous:
-            y = np.ascontiguousarray(y, dtype=DTYPE)
+            y = np.asarray(y, dtype=DTYPE, order="C")
 
         if sample_mask is None:
             sample_mask = np.ones((X.shape[0],), dtype=np.bool)
@@ -346,7 +361,8 @@ def recursive_partition(X, X_argsorted, y, sample_mask, depth,
         return self
 
     def predict(self, X):
-        out = np.empty((X.shape[0], self.value.shape[1]), dtype=np.float64)
+        out = np.empty((X.shape[0], self.value.shape[1], self.value.shape[2]),
+                       dtype=np.float64)
 
         _tree._predict_tree(X,
                             self.children,
@@ -430,6 +446,7 @@ def __init__(self, criterion,
         self.random_state = check_random_state(random_state)
 
         self.n_features_ = None
+        self.n_outputs_ = None
         self.classes_ = None
         self.n_classes_ = None
         self.find_split_ = _tree._find_best_split
@@ -445,7 +462,7 @@ def fit(self, X, y, sample_mask=None, X_argsorted=None):
         X : array-like of shape = [n_samples, n_features]
             The training input samples.
 
-        y : array-like, shape = [n_samples]
+        y : array-like, shape = [n_samples] or [n_samples, n_outputs]
             The target values (integers that correspond to classes in
             classification, real numbers in regression).
 
@@ -455,27 +472,42 @@ def fit(self, X, y, sample_mask=None, X_argsorted=None):
             Returns self.
         """
         # set min_samples_split sensibly
-        self.min_samples_split = max(self.min_samples_split, 2 *
-                self.min_samples_leaf)
+        self.min_samples_split = max(self.min_samples_split,
+                                     2 * self.min_samples_leaf)
 
         # Convert data
-        X = np.asarray(X, dtype=DTYPE, order='F')
+        X = np.asarray(X, dtype=DTYPE, order="F")
         n_samples, self.n_features_ = X.shape
 
         is_classification = isinstance(self, ClassifierMixin)
 
+        y = np.copy(y)
+        y = np.atleast_1d(y)
+        if y.ndim == 1:
+            y = y[:, np.newaxis]
+
+        self.classes_ = []
+        self.n_classes_ = []
+        self.n_outputs_ = y.shape[1]
+
         if is_classification:
-            self.classes_ = np.unique(y)
-            self.n_classes_ = self.classes_.shape[0]
-            criterion = CLASSIFICATION[self.criterion](self.n_classes_)
-            y = np.searchsorted(self.classes_, y)
+            for k in xrange(self.n_outputs_):
+                unique = np.unique(y[:, k])
+                self.classes_.append(unique)
+                self.n_classes_.append(unique.shape[0])
+                y[:, k] = np.searchsorted(unique, y[:, k])
 
         else:
-            self.classes_ = None
-            self.n_classes_ = 1
-            criterion = REGRESSION[self.criterion]()
+            self.classes_ = [None] * self.n_outputs_
+            self.n_classes_ = [1] * self.n_outputs_
+
+        y = np.asarray(y, dtype=DTYPE, order="C")
 
-        y = np.ascontiguousarray(y, dtype=DTYPE)
+        if is_classification:
+            criterion = CLASSIFICATION[self.criterion](self.n_outputs_,
+                                                       self.n_classes_)
+        else:
+            criterion = REGRESSION[self.criterion](self.n_outputs_)
 
         # Check parameters
         max_depth = np.inf if self.max_depth is None else self.max_depth
@@ -484,24 +516,18 @@ def fit(self, X, y, sample_mask=None, X_argsorted=None):
             if self.max_features == "auto":
                 if is_classification:
                     max_features = max(1, int(np.sqrt(self.n_features_)))
-
                 else:
                     max_features = self.n_features_
-
             elif self.max_features == "sqrt":
                 max_features = max(1, int(np.sqrt(self.n_features_)))
-
             elif self.max_features == "log2":
                 max_features = max(1, int(np.log2(self.n_features_)))
-
             else:
                 raise ValueError(
                     'Invalid value for max_features. Allowed string '
                     'values are "auto", "sqrt" or "log2".')
-
         elif self.max_features is None:
             max_features = self.n_features_
-
         else:
             max_features = self.max_features
 
@@ -520,7 +546,7 @@ def fit(self, X, y, sample_mask=None, X_argsorted=None):
             raise ValueError("max_features must be in (0, n_features]")
 
         # Build tree
-        self.tree_ = Tree(self.n_classes_, self.n_features_)
+        self.tree_ = Tree(self.n_classes_, self.n_features_, self.n_outputs_)
         self.tree_.build(X, y, criterion, max_depth,
                 self.min_samples_split, self.min_samples_leaf,
                 self.min_density, max_features, self.random_state,
@@ -547,7 +573,7 @@ def predict(self, X):
 
         Returns
         -------
-        y : array of shape = [n_samples]
+        y : array of shape = [n_samples] or [n_samples, n_outputs]
             The predicted classes, or the predict values.
         """
         X = array2d(X, dtype=DTYPE)
@@ -562,11 +588,20 @@ def predict(self, X):
                              " input n_features is %s "
                              % (self.n_features_, n_features))
 
+        P = self.tree_.predict(X)
+
         if isinstance(self, ClassifierMixin):
-            predictions = self.classes_.take(np.argmax(
-                self.tree_.predict(X), axis=1), axis=0)
+            predictions = np.zeros((n_samples, self.n_outputs_))
+
+            for k in xrange(self.n_outputs_):
+                predictions[:, k] = self.classes_[k].take(np.argmax(P[:, k],
+                                                                    axis=1),
+                                                          axis=0)
         else:
-            predictions = self.tree_.predict(X).ravel()
+            predictions = P[:, :, 0]
+
+        if self.n_outputs_ == 1:
+            predictions = predictions.reshape((n_samples, ))
 
         return predictions
 
@@ -691,7 +726,8 @@ def predict_proba(self, X):
 
         Returns
         -------
-        p : array of shape = [n_samples, n_classes]
+        p : array of shape = [n_samples, n_classes], or a list of n_outputs
+            such arrays if n_outputs > 1.
             The class probabilities of the input samples. Classes are ordered
             by arithmetical order.
         """
@@ -707,11 +743,21 @@ def predict_proba(self, X):
                              " input n_features is %s "
                              % (self.n_features_, n_features))
 
+        proba = []
         P = self.tree_.predict(X)
-        normalizer = P.sum(axis=1)[:, np.newaxis]
-        normalizer[normalizer == 0.0] = 1.0
-        P /= normalizer
-        return P
+
+        for k in xrange(self.n_outputs_):
+            P_k = P[:, k, :self.n_classes_[k]]
+            normalizer = P_k.sum(axis=1)[:, np.newaxis]
+            normalizer[normalizer == 0.0] = 1.0
+            P_k /= normalizer
+            proba.append(P_k)
+
+        if self.n_outputs_ == 1:
+            return proba[0]
+
+        else:
+            return proba
 
     def predict_log_proba(self, X):
         """Predict class log-probabilities of the input samples X.
@@ -723,11 +769,21 @@ def predict_log_proba(self, X):
 
         Returns
         -------
-        p : array of shape = [n_samples, n_classes]
+        p : array of shape = [n_samples, n_classes], or a list of n_outputs
+            such arrays if n_outputs > 1.
             The class log-probabilities of the input samples. Classes are
             ordered by arithmetical order.
         """
-        return np.log(self.predict_proba(X))
+        proba = self.predict_proba(X)
+
+        if self.n_outputs_ == 1:
+            return np.log(proba)
+
+        else:
+            for k in xrange(self.n_outputs_):
+                proba[k] = np.log(proba[k])
+
+            return proba
 
 
 class DecisionTreeRegressor(BaseDecisionTree, RegressorMixin):