BUG: restore score functionality in grid_search

GaelVaroquaux · GaelVaroquaux · commit 021a4e0d8ba9 · 2012-01-18T14:24:24.000+01:00
diff --git a/sklearn/grid_search.py b/sklearn/grid_search.py
@@ -370,8 +370,6 @@ def fit(self, X, y=None, **params):
             self.predict = best_estimator.predict
         if hasattr(best_estimator, 'predict_proba'):
             self.predict_proba = best_estimator.predict_proba
-        if hasattr(best_estimator, 'score'):
-            self.score_ = best_estimator.score
 
         # Store the computed scores
         # XXX: the name is too specific, it shouldn't have
@@ -383,8 +381,12 @@ def fit(self, X, y=None, **params):
         return self
 
     def score(self, X, y=None):
-        # This method is overridden during the fit if the best estimator
-        # found has a score function.
+        if hasattr(self.best_estimator, 'score'):
+             return self.best_estimator.score(X, y)
+        if self.score_func is None:
+            raise ValueError("No score function explicitly defined, "
+                             "and the estimator doesn't provide one %s"
+                             % self.best_estimator)
         y_predicted = self.predict(X)
         return self.score_func(y, y_predicted)
 
diff --git a/sklearn/tests/test_grid_search.py b/sklearn/tests/test_grid_search.py
@@ -42,12 +42,15 @@ def score(self, X=None, Y=None):
 def test_grid_search():
     """Test that the best estimator contains the right value for foo_param"""
     clf = MockClassifier()
-    cross_validation = GridSearchCV(clf, {'foo_param': [1, 2, 3]})
+    grid_search = GridSearchCV(clf, {'foo_param': [1, 2, 3]})
     # make sure it selects the smallest parameter in case of ties
-    assert_equal(cross_validation.fit(X, y).best_estimator_.foo_param, 2)
+    grid_search.fit(X, y)
+    assert_equal(grid_search.best_estimator_.foo_param, 2)
 
     for i, foo_i in enumerate([1, 2, 3]):
-        assert cross_validation.grid_scores_[i][0] == {'foo_param': foo_i}
+        assert grid_search.grid_scores_[i][0] == {'foo_param': foo_i}
+    # Smoke test the score:
+    grid_search.score(X, y)
 
 
 def test_grid_search_error():
@@ -101,6 +104,9 @@ def test_grid_search_sparse_score_func():
 
     assert_array_equal(y_pred, y_pred2)
     assert_equal(C, C2)
+    # Smoke test the score
+    #np.testing.assert_allclose(f1_score(cv.predict(X_[:180]), y[:180]),
+    #                        cv.score(X_[:180], y[:180]))
 
 
 class BrokenClassifier(BaseEstimator):
