issue #4931

preddy5 · preddy5 · commit b969bf6fa9c9 · 2015-11-02T11:00:49.000+05:30
diff --git a/doc/modules/neighbors.rst b/doc/modules/neighbors.rst
@@ -419,12 +419,14 @@ depends on a number of factors:
   a significant fraction of the total cost.  If very few query points
   will be required, brute force is better than a tree-based method.
 
-Currently, ``algorithm = 'auto'`` selects ``'kd_tree'`` if :math:`k < N/2` 
-and the ``'effective_metric_'`` is in the ``'VALID_METRICS'`` list of 
-``'kd_tree'``. It selects ``'ball_tree'`` if :math:`k < N/2` and the 
-``'effective_metric_'`` is not in the ``'VALID_METRICS'`` list of 
-``'kd_tree'``. It selects ``'brute'`` if :math:`k >= N/2`. This choice is based on the assumption that the number of query points is at least the 
-same order as the number of training points, and that ``leaf_size`` is 
+Currently, ``algorithm = 'auto'`` selects ``'kd_tree'`` if :math:`k < N/2`
+and the ``'effective_metric_'`` is in the ``'VALID_METRICS'`` list of
+``'kd_tree'``. It selects ``'ball_tree'`` if :math:`k < N/2` and the
+``'effective_metric_'`` is in the ``'VALID_METRICS'`` list of
+``'ball_tree'``. It selects ``'brute'`` if :math:`k < N/2` and the
+``'effective_metric_'`` is not in the ``'VALID_METRICS'`` list of
+``'kd_tree'`` or ``'ball_tree'``. It selects ``'brute'`` if :math:`k >= N/2`. This choice is based on the assumption that the number of query points is at least the
+same order as the number of training points, and that ``leaf_size`` is
 close to its default value of ``30``.
 
 Effect of ``leaf_size``
@@ -666,7 +668,7 @@ the :math:`m` nearest neighbors of a point :math:`q`. First, a top-down
 traversal is performed using a binary search to identify the leaf having the
 longest prefix match (maximum depth) with :math:`q`'s label after subjecting
 :math:`q` to the same hash functions. :math:`M >> m` points (total candidates)
-are extracted from the forest, moving up from the previously found maximum 
+are extracted from the forest, moving up from the previously found maximum
 depth towards the root synchronously across all trees in the bottom-up
 traversal. `M` is set to  :math:`cl` where :math:`c`, the number of candidates
 extracted from each tree, is a constant. Finally, the similarity of each of
diff --git a/sklearn/neighbors/base.py b/sklearn/neighbors/base.py
@@ -145,6 +145,8 @@ def _init_params(self, n_neighbors=None, radius=None,
                 alg_check = 'brute'
             else:
                 alg_check = 'ball_tree'
+                if metric not in VALID_METRICS[alg_check]:
+                    alg_check = 'brute'
         else:
             alg_check = algorithm
 
@@ -246,8 +248,10 @@ def _fit(self, X):
                     self.metric != 'precomputed'):
                 if self.effective_metric_ in VALID_METRICS['kd_tree']:
                     self._fit_method = 'kd_tree'
-                else:
+                elif self.effective_metric_ in VALID_METRICS['ball_tree']:
                     self._fit_method = 'ball_tree'
+                else:
+                    self._fit_method = 'brute'
             else:
                 self._fit_method = 'brute'
 
diff --git a/sklearn/neighbors/tests/test_neighbors.py b/sklearn/neighbors/tests/test_neighbors.py
@@ -975,6 +975,17 @@ def test_callable_metric():
     assert_array_almost_equal(dist1, dist2)
 
 
+def test_algo_auto_metrics():
+    X = rng.rand(12, 3)
+    Xcsr = csr_matrix(X)
+    nn = neighbors.NearestNeighbors(n_neighbors=3, algorithm='auto',
+                                    metric='cosine').fit(X)
+    assert_true(nn._fit_method, 'brute')
+    nn = neighbors.NearestNeighbors(n_neighbors=3, algorithm='auto',
+                                    metric='cosine').fit(Xcsr)
+    assert_true(nn._fit_method, 'brute')
+
+
 def test_metric_params_interface():
     assert_warns(DeprecationWarning, neighbors.KNeighborsClassifier,
                  metric='wminkowski', w=np.ones(10))
