Merge pull request scikit-learn#4894 from tw991/rf

fix dtype transform problem in KNN and RandomForest

Author: agramfort

sklearn/ensemble/forest.py

@@ -414,10 +414,12 @@ def _validate_y_class_weight(self, y):
         self.classes_ = []
         self.n_classes_ = []
 
+        y_store_unique_indices = np.zeros(y.shape, dtype=np.int)
         for k in range(self.n_outputs_):
-            classes_k, y[:, k] = np.unique(y[:, k], return_inverse=True)
+            classes_k, y_store_unique_indices[:, k] = np.unique(y[:, k], return_inverse=True)
             self.classes_.append(classes_k)
             self.n_classes_.append(classes_k.shape[0])
+        y = y_store_unique_indices
 
         if self.class_weight is not None:
             valid_presets = ('auto', 'balanced', 'balanced_subsample', 'subsample', 'auto')

sklearn/ensemble/tests/test_forest.py

@@ -971,3 +971,13 @@ def test_warm_start_oob():
         yield check_warm_start_oob, name
     for name in FOREST_REGRESSORS:
         yield check_warm_start_oob, name
+
+
+def test_dtype_convert():
+    classifier = RandomForestClassifier()
+    CLASSES = 15
+    X = np.eye(CLASSES)
+    y = [ch for ch in 'ABCDEFGHIJKLMNOPQRSTU'[:CLASSES]]
+
+    result = classifier.fit(X, y).predict(X)
+    assert_array_equal(result, y)

sklearn/neighbors/tests/test_neighbors.py

@@ -1082,3 +1082,13 @@ def test_include_self_neighbors_graph():
         X, 5.0, include_self=False).A
     assert_array_equal(rng, [[1., 1.], [1., 1.]])
     assert_array_equal(rng_not_self, [[0., 1.], [1., 0.]])
+
+
+def test_dtype_convert():
+    classifier = neighbors.KNeighborsClassifier(n_neighbors=1)
+    CLASSES = 15
+    X = np.eye(CLASSES)
+    y = [ch for ch in 'ABCDEFGHIJKLMNOPQRSTU'[:CLASSES]]
+
+    result = classifier.fit(X, y).predict(X)
+    assert_array_equal(result, y)

sklearn/tree/tree.py

@@ -166,10 +166,12 @@ def fit(self, X, y, sample_weight=None, check_input=True):
             if self.class_weight is not None:
                 y_original = np.copy(y)
 
+            y_store_unique_indices = np.zeros(y.shape, dtype=np.int)
             for k in range(self.n_outputs_):
-                classes_k, y[:, k] = np.unique(y[:, k], return_inverse=True)
+                classes_k, y_store_unique_indices[:, k] = np.unique(y[:, k], return_inverse=True)
                 self.classes_.append(classes_k)
                 self.n_classes_.append(classes_k.shape[0])
+            y = y_store_unique_indices
 
             if self.class_weight is not None:
                 expanded_class_weight = compute_sample_weight(