FIX check_methods_subset_invariance where estimator produces sparse output (scikit-learn#11173)

Author: jnothman

sklearn/utils/estimator_checks.py

@@ -676,7 +676,7 @@ def check_fit2d_predict1d(name, estimator_orig):
                                  getattr(estimator, method), X[0])
 
 
-def _apply_func(func, X):
+def _apply_on_subsets(func, X):
     # apply function on the whole set and on mini batches
     result_full = func(X)
     n_features = X.shape[1]
@@ -687,6 +687,9 @@ def _apply_func(func, X):
         result_full = result_full[0]
         result_by_batch = list(map(lambda x: x[0], result_by_batch))
 
+    if sparse.issparse(result_full):
+        result_full = result_full.A
+        result_by_batch = [x.A for x in result_by_batch]
     return np.ravel(result_full), np.ravel(result_by_batch)
 
 
@@ -722,7 +725,7 @@ def check_methods_subset_invariance(name, estimator_orig):
             raise SkipTest(msg)
 
         if hasattr(estimator, method):
-            result_full, result_by_batch = _apply_func(
+            result_full, result_by_batch = _apply_on_subsets(
                 getattr(estimator, method), X)
             assert_allclose(result_full, result_by_batch,
                             atol=1e-7, err_msg=msg)

sklearn/utils/tests/test_estimator_checks.py

@@ -161,6 +161,21 @@ def predict(self, X):
         return np.zeros(X.shape[0])
 
 
+class SparseTransformer(BaseEstimator):
+    def fit(self, X, y=None):
+        self.X_shape_ = check_array(X).shape
+        return self
+
+    def fit_transform(self, X, y=None):
+        return self.fit(X, y).transform(X)
+
+    def transform(self, X):
+        X = check_array(X)
+        if X.shape[1] != self.X_shape_[1]:
+            raise ValueError('Bad number of features')
+        return sp.csr_matrix(X)
+
+
 def test_check_estimator():
     # tests that the estimator actually fails on "bad" estimators.
     # not a complete test of all checks, which are very extensive.
@@ -235,6 +250,9 @@ def test_check_estimator():
         sys.stdout = old_stdout
     assert_true(msg in string_buffer.getvalue())
 
+    # non-regression test for estimators transforming to sparse data
+    check_estimator(SparseTransformer())
+
     # doesn't error on actual estimator
     check_estimator(AdaBoostClassifier)
     check_estimator(AdaBoostClassifier())