warning for PCA with sparse input (scikit-learn#7649)

Author: Giorgio Patrini

sklearn/decomposition/pca.py

@@ -15,6 +15,7 @@
 import numpy as np
 from scipy import linalg
 from scipy.special import gammaln
+from scipy.sparse import issparse
 
 from ..externals import six
 
@@ -116,6 +117,9 @@ class PCA(_BasePCA):
     It can also use the scipy.sparse.linalg ARPACK implementation of the
     truncated SVD.
 
+    Notice that this class does not support sparse input. See
+    :ref:`<TruncatedSVD>` for an alternative with sparse data.
+
     Read more in the :ref:`User Guide <PCA>`.
 
     Parameters
@@ -332,6 +336,13 @@ def fit_transform(self, X, y=None):
 
     def _fit(self, X):
         """Dispatch to the right submethod depending on the chosen solver."""
+
+        # Raise an error for sparse input.
+        # This is more informative than the generic one raised by check_array.
+        if issparse(X):
+            raise TypeError('PCA does not support sparse input. See '
+                            'TruncatedSVD for a possible alternative.')
+
         X = check_array(X, dtype=[np.float64], ensure_2d=True,
                         copy=self.copy)
 

sklearn/decomposition/tests/test_pca.py

@@ -1,4 +1,5 @@
 import numpy as np
+import scipy as sp
 from itertools import product
 
 from sklearn.utils.testing import assert_almost_equal
@@ -508,3 +509,15 @@ def fit_deprecated(X):
     assert_warns_message(DeprecationWarning, depr_message, fit_deprecated, X)
     Y_pca = PCA(svd_solver='randomized', random_state=0).fit_transform(X)
     assert_array_almost_equal(Y, Y_pca)
+
+
+def test_pca_spase_input():
+
+    X = np.random.RandomState(0).rand(5, 4)
+    X = sp.sparse.csr_matrix(X)
+    assert(sp.sparse.issparse(X))
+
+    for svd_solver in solver_list:
+        pca = PCA(n_components=3, svd_solver=svd_solver)
+
+        assert_raises(TypeError, pca.fit, X)

sklearn/decomposition/truncated_svd.py

@@ -35,7 +35,7 @@ class TruncatedSVD(BaseEstimator, TransformerMixin):
     returned by the vectorizers in sklearn.feature_extraction.text. In that
     context, it is known as latent semantic analysis (LSA).
 
-    This estimator supports two algorithm: a fast randomized SVD solver, and
+    This estimator supports two algorithms: a fast randomized SVD solver, and
     a "naive" algorithm that uses ARPACK as an eigensolver on (X * X.T) or
     (X.T * X), whichever is more efficient.