fix type_of_target for csr_matrices

alk-lbinet · alk-lbinet · commit f2124005c897 · 2019-09-01T14:55:53.000+02:00
diff --git a/sklearn/utils/multiclass.py b/sklearn/utils/multiclass.py
@@ -16,7 +16,7 @@
 
 import numpy as np
 
-from .validation import check_array, _assert_all_finite
+from .validation import check_array, _assert_all_finite, assert_all_finite
 
 
 def _unique_multiclass(y):
@@ -247,11 +247,14 @@ def type_of_target(y):
     if is_multilabel(y):
         return 'multilabel-indicator'
 
-    try:
-        y = np.asarray(y)
-    except ValueError:
-        # Known to fail in numpy 1.3 for array of arrays
-        return 'unknown'
+    if not issparse(y):
+        # calling np.asarray on sparse matrix has unexpected behavior
+        # https://github.com/numpy/numpy/issues/14221
+        try:
+            y = np.asarray(y)
+        except ValueError:
+            # Known to fail in numpy 1.3 for array of arrays
+            return 'unknown'
 
     # The old sequence of sequences format
     try:
@@ -266,9 +269,13 @@ def type_of_target(y):
         pass
 
     # Invalid inputs
-    if y.ndim > 2 or (y.dtype == object and len(y) and
-                      not isinstance(y.flat[0], str)):
+    if y.ndim > 2:
+        return 'unknown'
+    if not issparse(y) and y.dtype == object \
+            and not isinstance(y.flat[0], str):
         return 'unknown'  # [[[1, 2]]] or [obj_1] and not ["label_1"]
+    if issparse(y) and y.dtype == object and not isinstance(y.data[0], str):
+        return 'unknown'  # [[[1, 2]]] or [obj_1] and not ["label_1"] (sparse)
 
     if y.ndim == 2 and y.shape[1] == 0:
         return 'unknown'  # [[]]
@@ -279,13 +286,22 @@ def type_of_target(y):
         suffix = ""  # [1, 2, 3] or [[1], [2], [3]]
 
     # check float and contains non-integer float values
-    if y.dtype.kind == 'f' and np.any(y != y.astype(int)):
-        # [.1, .2, 3] or [[.1, .2, 3]] or [[1., .2]] and not [1., 2., 3.]
-        _assert_all_finite(y)
-        return 'continuous' + suffix
-
-    if (len(np.unique(y)) > 2) or (y.ndim >= 2 and len(y[0]) > 1):
-        return 'multiclass' + suffix  # [1, 2, 3] or [[1., 2., 3]] or [[1, 2]]
+    if y.dtype.kind == 'f':
+        if not issparse(y) and np.any(y != y.astype(int)):
+            # [.1, .2, 3] or [[.1, .2, 3]] or [[1., .2]] and not [1., 2., 3.]
+            _assert_all_finite(y)
+            return 'continuous' + suffix
+        if issparse(y) and np.any(y.data != y.data.astype(int)):
+            # [.1, .2, 3] or [[.1, .2, 3]] or [[1., .2]] and not [1., 2., 3.]
+            assert_all_finite(y)
+            return 'continuous' + suffix
+
+    if len(np.unique(y)) > 2:
+        return 'multiclass' + suffix  # [1, 2, 3] or [[1., 2., 3]]
+    if not issparse(y) and y.ndim >= 2 and len(y[0]) > 1:
+        return 'multiclass' + suffix  # [[1, 2]] or [[0],[1]]
+    if issparse(y) and y.ndim >= 2:
+        return 'multiclass' + suffix  # [[1, 2]]
     else:
         return 'binary'  # [1, 2] or [["a"], ["b"]]
 
diff --git a/sklearn/utils/tests/test_multiclass.py b/sklearn/utils/tests/test_multiclass.py
@@ -81,6 +81,11 @@ def __array__(self, dtype=None):
         np.array([[1, 0, 2, 2], [1, 4, 2, 4]], dtype=np.uint8),
         np.array([[1, 0, 2, 2], [1, 4, 2, 4]], dtype=np.float),
         np.array([[1, 0, 2, 2], [1, 4, 2, 4]], dtype=np.float32),
+        csr_matrix(np.array([[1, 0, 2, 2], [1, 4, 2, 4]])),
+        csr_matrix(np.array([[1, 0, 2, 2], [1, 4, 2, 4]]), dtype=np.int8),
+        csr_matrix(np.array([[1, 0, 2, 2], [1, 4, 2, 4]]), dtype=np.uint8),
+        csr_matrix(np.array([[1, 0, 2, 2], [1, 4, 2, 4]]), dtype=np.float),
+        csr_matrix(np.array([[1, 0, 2, 2], [1, 4, 2, 4]]), dtype=np.float32),
         np.array([['a', 'b'], ['c', 'd']]),
         np.array([['a', 'b'], ['c', 'd']]),
         np.array([['a', 'b'], ['c', 'd']], dtype=object),
@@ -119,6 +124,9 @@ def __array__(self, dtype=None):
         np.array([[0, .5], [.5, 0]]),
         np.array([[0, .5], [.5, 0]], dtype=np.float32),
         np.array([[0, .5]]),
+        csr_matrix(np.array([[0, .5], [.5, 0]])),
+        csr_matrix(np.array([[0, .5], [.5, 0]]), dtype=np.float32),
+        csr_matrix(np.array([[0, .5]])),
     ],
     'unknown': [
         [[]],
