ENH add inverse_transform in *SparsePCA (#23905)

glemaitre · web-flow · commit 01fcf8a0acc7 · 2022-07-19T09:44:35.000-04:00
diff --git a/doc/whats_new/v1.2.rst b/doc/whats_new/v1.2.rst
@@ -125,6 +125,11 @@ Changelog
   its memory footprint and runtime.
   :pr:`22268` by :user:`MohamedBsh <Bsh>`.
 
+- |Enhancement| :class:`decomposition.SparsePCA` and
+  :class:`decomposition.MiniBatchSparsePCA` now implements an `inverse_transform`
+  function.
+  :pr:`23905` by :user:`Guillaume Lemaitre <glemaitre>`.
+
 :mod:`sklearn.decomposition`
 ............................
 
diff --git a/sklearn/decomposition/_sparse_pca.py b/sklearn/decomposition/_sparse_pca.py
@@ -8,7 +8,7 @@
 
 from ..utils import check_random_state
 from ..utils._param_validation import Hidden, Interval, StrOptions
-from ..utils.validation import check_is_fitted
+from ..utils.validation import check_array, check_is_fitted
 from ..linear_model import ridge_regression
 from ..base import BaseEstimator, TransformerMixin, _ClassNamePrefixFeaturesOutMixin
 from ._dict_learning import dict_learning, MiniBatchDictionaryLearning
@@ -115,6 +115,29 @@ def transform(self, X):
 
         return U
 
+    def inverse_transform(self, X):
+        """Transform data from the latent space to the original space.
+
+        This inversion is an approximation due to the loss of information
+        induced by the forward decomposition.
+
+        .. versionadded:: 1.2
+
+        Parameters
+        ----------
+        X : ndarray of shape (n_samples, n_components)
+            Data in the latent space.
+
+        Returns
+        -------
+        X_original : ndarray of shape (n_samples, n_features)
+            Reconstructed data in the original space.
+        """
+        check_is_fitted(self)
+        X = check_array(X)
+
+        return (X @ self.components_) + self.mean_
+
     @property
     def _n_features_out(self):
         """Number of transformed output features."""
diff --git a/sklearn/decomposition/tests/test_sparse_pca.py b/sklearn/decomposition/tests/test_sparse_pca.py
@@ -313,3 +313,38 @@ def test_spca_early_stopping(global_random_seed):
         max_iter=100, tol=1e-6, max_no_improvement=100, random_state=global_random_seed
     ).fit(X)
     assert model_early_stopped.n_iter_ < model_not_early_stopped.n_iter_
+
+
+def test_sparse_pca_inverse_transform():
+    """Check that `inverse_transform` in `SparsePCA` and `PCA` are similar."""
+    rng = np.random.RandomState(0)
+    n_samples, n_features = 10, 5
+    X = rng.randn(n_samples, n_features)
+
+    n_components = 2
+    spca = SparsePCA(
+        n_components=n_components, alpha=1e-12, ridge_alpha=1e-12, random_state=0
+    )
+    pca = PCA(n_components=n_components, random_state=0)
+    X_trans_spca = spca.fit_transform(X)
+    X_trans_pca = pca.fit_transform(X)
+    assert_allclose(
+        spca.inverse_transform(X_trans_spca), pca.inverse_transform(X_trans_pca)
+    )
+
+
+@pytest.mark.parametrize("SPCA", [SparsePCA, MiniBatchSparsePCA])
+def test_transform_inverse_transform_round_trip(SPCA):
+    """Check the `transform` and `inverse_transform` round trip with no loss of
+    information.
+    """
+    rng = np.random.RandomState(0)
+    n_samples, n_features = 10, 5
+    X = rng.randn(n_samples, n_features)
+
+    n_components = n_features
+    spca = SPCA(
+        n_components=n_components, alpha=1e-12, ridge_alpha=1e-12, random_state=0
+    )
+    X_trans_spca = spca.fit_transform(X)
+    assert_allclose(spca.inverse_transform(X_trans_spca), X)
