scikit-learn · lorentzenchr · Jun 17, 2021 · Jun 26, 2020 · Jun 29, 2020 · Jun 29, 2020
diff --git a/doc/whats_new/v1.0.rst b/doc/whats_new/v1.0.rst
@@ -358,11 +358,17 @@ Changelog
   LinearRegression was deprecated in:
   :pr:`17743` by :user:`Maria Telenczuk <maikia>` and
   :user:`Alexandre Gramfort <agramfort>`.
-  Ridge, RidgeClassifier, RidgeCV or RidgeClassifierCV were deprecated in:
+  The ``normalize`` parameter in Ridge, RidgeClassifier, RidgeCV or
+  RidgeClassifierCV were deprecated and will be removed in 1.2.
   :pr:`17772` by :user:`Maria Telenczuk <maikia>` and
   :user:`Alexandre Gramfort <agramfort>`.
-  BayesianRidge, ARDRegression were deprecated in:
+  Same for BayesianRidge, ARDRegression in:
   :pr:`17746` by :user:`Maria Telenczuk <maikia>`.
+  The ``normalize`` parameter of :class:`linear_model.OrthogonalMatchingPursuit`
+  :class:`linear_model.OrthogonalMatchingPursuitCV` will default to
+  False in 1.2 and will be removed in 1.4.
+  :pr:`17750` by :user:`Maria Telenczuk <maikia>` and
+  :user:`Alexandre Gramfort <agramfort>`.
 
 - |Fix| `sample_weight` are now fully taken into account in linear models
   when `normalize=True` for both feature centering and feature

diff --git a/examples/linear_model/plot_omp.py b/examples/linear_model/plot_omp.py
@@ -41,7 +41,8 @@
 plt.stem(idx, w[idx], use_line_collection=True)
 
 # plot the noise-free reconstruction
-omp = OrthogonalMatchingPursuit(n_nonzero_coefs=n_nonzero_coefs)
+omp = OrthogonalMatchingPursuit(n_nonzero_coefs=n_nonzero_coefs,
+                                normalize=False)
 omp.fit(X, y)
 coef = omp.coef_
 idx_r, = coef.nonzero()
@@ -60,7 +61,7 @@
 plt.stem(idx_r, coef[idx_r], use_line_collection=True)
 
 # plot the noisy reconstruction with number of non-zeros set by CV
-omp_cv = OrthogonalMatchingPursuitCV()
+omp_cv = OrthogonalMatchingPursuitCV(normalize=False)
 omp_cv.fit(X, y_noisy)
 coef = omp_cv.coef_
 idx_r, = coef.nonzero()

diff --git a/sklearn/linear_model/_omp.py b/sklearn/linear_model/_omp.py
@@ -13,7 +13,7 @@
 from scipy.linalg.lapack import get_lapack_funcs
 from joblib import Parallel
 
-from ._base import LinearModel, _pre_fit
+from ._base import LinearModel, _pre_fit, _deprecate_normalize
 from ..base import RegressorMixin, MultiOutputMixin
 from ..utils import as_float_array, check_array
 from ..utils.fixes import delayed
@@ -616,6 +616,10 @@ class OrthogonalMatchingPursuit(MultiOutputMixin, RegressorMixin, LinearModel):
         :class:`~sklearn.preprocessing.StandardScaler` before calling ``fit``
         on an estimator with ``normalize=False``.
 
+        .. deprecated:: 1.0
+            ``normalize`` was deprecated in version 1.0. It will default
+            to False in 1.2 and be removed in 1.4.
+
     precompute : 'auto' or bool, default='auto'
         Whether to use a precomputed Gram and Xy matrix to speed up
         calculations. Improves performance when :term:`n_targets` or
@@ -648,7 +652,7 @@ class OrthogonalMatchingPursuit(MultiOutputMixin, RegressorMixin, LinearModel):
     >>> from sklearn.linear_model import OrthogonalMatchingPursuit
     >>> from sklearn.datasets import make_regression
     >>> X, y = make_regression(noise=4, random_state=0)
-    >>> reg = OrthogonalMatchingPursuit().fit(X, y)
+    >>> reg = OrthogonalMatchingPursuit(normalize=False).fit(X, y)
     >>> reg.score(X, y)
     0.9991...
     >>> reg.predict(X[:1,])
@@ -683,7 +687,7 @@ def __init__(
         n_nonzero_coefs=None,
         tol=None,
         fit_intercept=True,
-        normalize=True,
+        normalize="deprecated",
         precompute="auto",
     ):
         self.n_nonzero_coefs = n_nonzero_coefs
@@ -709,11 +713,15 @@ def fit(self, X, y):
         self : object
             returns an instance of self.
         """
+        _normalize = _deprecate_normalize(
+            self.normalize, default=True, estimator_name=self.__class__.__name__
+        )
+
         X, y = self._validate_data(X, y, multi_output=True, y_numeric=True)
         n_features = X.shape[1]
 
         X, y, X_offset, y_offset, X_scale, Gram, Xy = _pre_fit(
-            X, y, None, self.precompute, self.normalize, self.fit_intercept, copy=True
+            X, y, None, self.precompute, _normalize, self.fit_intercept, copy=True
         )
 
         if y.ndim == 1:
@@ -797,6 +805,10 @@ def _omp_path_residues(
         :class:`~sklearn.preprocessing.StandardScaler` before calling ``fit``
         on an estimator with ``normalize=False``.
 
+        .. deprecated:: 1.0
+            ``normalize`` was deprecated in version 1.0. It will default
+            to False in 1.2 and be removed in 1.4.
+
     max_iter : int, default=100
         Maximum numbers of iterations to perform, therefore maximum features
         to include. 100 by default.
@@ -872,6 +884,10 @@ class OrthogonalMatchingPursuitCV(RegressorMixin, LinearModel):
         :class:`~sklearn.preprocessing.StandardScaler` before calling ``fit``
         on an estimator with ``normalize=False``.
 
+        .. deprecated:: 1.0
+            ``normalize`` was deprecated in version 1.0. It will default
+            to False in 1.2 and be removed in 1.4.
+
     max_iter : int, default=None
         Maximum numbers of iterations to perform, therefore maximum features
         to include. 10% of ``n_features`` but at least 5 if available.
@@ -929,7 +945,7 @@ class OrthogonalMatchingPursuitCV(RegressorMixin, LinearModel):
     >>> from sklearn.datasets import make_regression
     >>> X, y = make_regression(n_features=100, n_informative=10,
     ...                        noise=4, random_state=0)
-    >>> reg = OrthogonalMatchingPursuitCV(cv=5).fit(X, y)
+    >>> reg = OrthogonalMatchingPursuitCV(cv=5, normalize=False).fit(X, y)
     >>> reg.score(X, y)
     0.9991...
     >>> reg.n_nonzero_coefs_
@@ -956,7 +972,7 @@ def __init__(
         *,
         copy=True,
         fit_intercept=True,
-        normalize=True,
+        normalize="deprecated",
         max_iter=None,
         cv=None,
         n_jobs=None,
@@ -986,6 +1002,11 @@ def fit(self, X, y):
         self : object
             returns an instance of self.
         """
+
+        _normalize = _deprecate_normalize(
+            self.normalize, default=True, estimator_name=self.__class__.__name__
+        )
+
         X, y = self._validate_data(
             X, y, y_numeric=True, ensure_min_features=2, estimator=self
         )
@@ -1004,7 +1025,7 @@ def fit(self, X, y):
                 y[test],
                 self.copy,
                 self.fit_intercept,
-                self.normalize,
+                _normalize,
                 max_iter,
             )
             for train, test in cv.split(X)
@@ -1019,7 +1040,7 @@ def fit(self, X, y):
         omp = OrthogonalMatchingPursuit(
             n_nonzero_coefs=best_n_nonzero_coefs,
             fit_intercept=self.fit_intercept,
-            normalize=self.normalize,
+            normalize=_normalize,
         )
         omp.fit(X, y)
         self.coef_ = omp.coef_

diff --git a/sklearn/linear_model/tests/test_omp.py b/sklearn/linear_model/tests/test_omp.py
@@ -35,6 +35,30 @@
 # and y (n_samples, 3)
 
 
+# FIXME: 'normalize' to set to False in 1.2 and removed in 1.4
+@pytest.mark.parametrize(
+    "OmpModel", [OrthogonalMatchingPursuit, OrthogonalMatchingPursuitCV]
+)
+@pytest.mark.parametrize(
+    "normalize, n_warnings", [(True, 0), (False, 0), ("deprecated", 1)]
+)
+def test_assure_warning_when_normalize(OmpModel, normalize, n_warnings):
+    # check that we issue a FutureWarning when normalize was set
+    rng = check_random_state(0)
+    n_samples = 200
+    n_features = 2
+    X = rng.randn(n_samples, n_features)
+    X[X < 0.1] = 0.0
+    y = rng.rand(n_samples)
+
+    model = OmpModel(normalize=normalize)
+    with pytest.warns(None) as record:
+        model.fit(X, y)
+
+    record = [r for r in record if r.category == FutureWarning]
+    assert len(record) == n_warnings
+
+
 def test_correct_shapes():
     assert orthogonal_mp(X, y[:, 0], n_nonzero_coefs=5).shape == (n_features,)
     assert orthogonal_mp(X, y, n_nonzero_coefs=5).shape == (n_features, 3)
@@ -125,6 +149,8 @@ def test_orthogonal_mp_gram_readonly():
     assert_array_almost_equal(gamma[:, 0], gamma_gram, decimal=2)
 
 
+# FIXME: 'normalize' to be removed in 1.4
+@pytest.mark.filterwarnings("ignore:The default of 'normalize'")
 def test_estimator():
     omp = OrthogonalMatchingPursuit(n_nonzero_coefs=n_nonzero_coefs)
     omp.fit(X, y[:, 0])
@@ -211,6 +237,8 @@ def test_omp_return_path_prop_with_gram():
     assert_array_almost_equal(path[:, :, -1], last)
 
 
+# FIXME: 'normalize' to be removed in 1.4
+@pytest.mark.filterwarnings("ignore:The default of 'normalize'")
 def test_omp_cv():
     y_ = y[:, 0]
     gamma_ = gamma[:, 0]
@@ -227,6 +255,8 @@ def test_omp_cv():
     assert_array_almost_equal(ompcv.coef_, omp.coef_)
 
 
+# FIXME: 'normalize' to be removed in 1.4
+@pytest.mark.filterwarnings("ignore:The default of 'normalize'")
 def test_omp_reaches_least_squares():
     # Use small simple data; it's a sanity check but OMP can stop early
     rng = check_random_state(0)

diff --git a/sklearn/tests/test_docstring_parameters.py b/sklearn/tests/test_docstring_parameters.py
@@ -241,6 +241,13 @@ def test_fit_docstring_attributes(name, Estimator):
         # default="auto" raises an error with the shape of `X`
         est.set_params(n_components=2)
 
+    # FIXME: TO BE REMOVED in 1.4 (avoid FutureWarning)
+    if Estimator.__name__ in (
+        "OrthogonalMatchingPursuit",
+        "OrthogonalMatchingPursuitCV",
+    ):
+        est.set_params(normalize=False)
+
     # FIXME: TO BE REMOVED for 1.1 (avoid FutureWarning)
     if Estimator.__name__ == "NMF":
         est.set_params(init="nndsvda")