scikit-learn
diff --git a/‎sklearn/feature_selection/sequential_feature_selector.py
Lines changed: 7 additions & 6 deletions b/‎sklearn/feature_selection/sequential_feature_selector.py
Lines changed: 7 additions & 6 deletions
diff --git a/‎sklearn/feature_selection/tests/test_sequential_feature_selector.py
Lines changed: 15 additions & 6 deletions b/‎sklearn/feature_selection/tests/test_sequential_feature_selector.py
Lines changed: 15 additions & 6 deletions
@@ -82,8 +82,8 @@ class SequentialFeatureSelector(BaseEstimator, MetaEstimatorMixin,
 
     Attributes
     ----------
-    feature_subset_idx_ : array-like, shape = [n_predictions]
-        Feature Indices of the selected feature subsets.
+    support_ : array of shape [n_features]
+        The mask of selected features.
 
     score_ : float
         Cross validation average score of the selected subset.
@@ -221,7 +221,7 @@ def fit(self, X, y):
             self.subsets_[k] = {
                 'feature_subset_idx': k_idx,
                 'cv_scores': k_score,
-                'avg_score': np.nanmean(k_score)
+                'avg_score': np.mean(k_score)
                 }
 
         best_subset = None
@@ -266,7 +266,8 @@ def fit(self, X, y):
             k_score = max_score
             k_idx = self.subsets_[best_subset]['feature_subset_idx']
 
-        self.feature_subset_idx_ = k_idx
+        self.support_ = k_idx
+        self.support_ = self._get_support_mask()
         self.score_ = k_score
         return self
 
@@ -324,8 +325,8 @@ def _exclusion(self, feature_set, X, y, estimator, fixed_feature=None):
         return res
 
     def _get_support_mask(self):
-        check_is_fitted(self, 'feature_subset_idx_')
+        check_is_fitted(self, 'support_')
         mask = np.zeros((self._n_features,), dtype=np.bool)
         # list to avoid IndexError in old NumPy versions
-        mask[list(self.feature_subset_idx_)] = True
+        mask[list(self.support_)] = True
         return mask
@@ -3,6 +3,7 @@
 """
 import numpy as np
 from numpy.testing import assert_almost_equal
+from numpy.testing import assert_array_equal
 from sklearn.utils.testing import assert_raise_message
 from sklearn.neighbors import KNeighborsClassifier
 from sklearn.linear_model import LinearRegression
@@ -37,7 +38,8 @@ def test_run_default():
     knn = KNeighborsClassifier()
     sfs = SFS(estimator=knn)
     sfs.fit(X, y)
-    assert sfs.feature_subset_idx_ == (3, )
+    assert_array_equal(sfs.support_,
+                       np.array([False, False, False, True]))
 
 
 def test_kfeatures_type_1():
@@ -163,7 +165,8 @@ def test_knn_option_sbs():
                forward=False,
                cv=4)
     sfs3 = sfs3.fit(X, y)
-    assert sfs3.feature_subset_idx_ == (1, 2, 3)
+    assert_array_equal(sfs3.support_,
+                       np.array([False, True, True, True]))
 
 
 def test_knn_option_sfs_tuplerange():
@@ -177,7 +180,8 @@ def test_knn_option_sfs_tuplerange():
                cv=4)
     sfs4 = sfs4.fit(X, y)
     assert round(sfs4.score_, 3) == 0.967
-    assert sfs4.feature_subset_idx_ == (0, 2, 3)
+    assert_array_equal(sfs4.support_,
+                       np.array([True, False, True, True]))
 
 
 def test_knn_scoring_metric():
@@ -217,7 +221,7 @@ def test_regression():
                 forward=True,
                 cv=10)
     sfs_r = sfs_r.fit(X, y)
-    assert len(sfs_r.feature_subset_idx_) == 13
+    assert sum(sfs_r.support_) == 13
     assert round(sfs_r.score_, 4) == 0.2001
 
 
@@ -233,7 +237,7 @@ def test_regression_in_tuplerange_forward():
                 forward=True,
                 cv=10)
     sfs_r = sfs_r.fit(X, y)
-    assert len(sfs_r.feature_subset_idx_) == 9
+    assert sum(sfs_r.support_) == 9
     assert round(sfs_r.score_, 4) == 0.2991, sfs_r.score_
 
 
@@ -252,7 +256,12 @@ def test_regression_in_tuplerange_backward():
                 cv=10)
 
     sfs_r = sfs_r.fit(X, y)
-    assert len(sfs_r.feature_subset_idx_) == 5
+    print(sfs_r.support_)
+    print(type(sfs_r.support_))
+    assert_array_equal(sfs_r.support_,
+                       np.array([False, False, False, False,
+                                 True, False, False,  True,
+                                 True, False, True, False, True]))
 
 
 def test_transform_not_fitted():