Densify one column at a time to do slicing, revert changes to sparsef…

…uncs_fast

sklearn/neighbors/classification.py

@@ -13,7 +13,6 @@
 
 from scipy import stats
 from sklearn.utils.extmath import weighted_mode
-from sklearn.utils.sparsefuncs_fast import csr_row_mode
 
 from sklearn.neighbors.base import \
     _check_weights, _get_weights, \
@@ -170,28 +169,13 @@ def predict(self, X):
             indptr = array.array('i', [0])
 
             for k, classes_k in enumerate(classes_):
-                # Using _y[neigh_ind, k] is not supported with scipy <0.13
-                # so we recreate fancy indexing using numpy functions
-                _y.sum_duplicates()
-                _y_data_k = _y.data[_y.indptr[k]:_y.indptr[k+1]]
-                _y_indices_k = _y.indices[_y.indptr[k]:_y.indptr[k+1]]
-
-                # Find the neigh_ind in _y.data using _y.indices as a guide
-                data_index = np.searchsorted(_y_indices_k, neigh_ind)
-                data_index[data_index == _y_data_k.shape[0]] = 0
-                if _y_data_k.size == 0:
-                    neigh_lbls_k = np.zeros(shape=data_index.shape)
-                else:
-                    neigh_lbls_k = _y_data_k[data_index]
-                    # Replace incorrect nonzero elements with correct zeros
-                    neigh_lbls_k[_y_indices_k[data_index] != neigh_ind] = 0
+                neigh_lbls_k = _y.getcol(k).toarray().ravel()[neigh_ind]
 
                 if weights is None:
-                    mode = csr_row_mode(sp.csr_matrix(neigh_lbls_k))
+                    mode, _ = stats.mode(neigh_lbls_k,  axis=1)
                     mode = sp.csc_matrix(mode, dtype=np.intp)
                 else:
-                    mode = mode = csr_row_mode(sp.csr_matrix(neigh_lbls_k),
-                                               weights)
+                    mode, _ = weighted_mode(neigh_lbls_k, weights, axis=1)
                     mode = sp.csc_matrix(mode, dtype=np.intp)
 
                 data.extend(mode.data)