Fix issue-#85: development branch

autosklearn/ensembles/ensemble_selection.py

@@ -156,9 +156,9 @@ def _calculate_weights(self):
     def _sorted_initialization(self, predictions, labels, n_best):
         perf = np.zeros([predictions.shape[0]])
 
-        for i, p in enumerate(predictions):
-            perf[i] = calculate_score(labels, predictions, self.task_type,
-                                      self.metric, predictions.shape[1])
+        for idx, prediction in enumerate(predictions):
+            perf[idx] = calculate_score(labels, prediction, self.task_type,
+                                        self.metric, predictions.shape[1])
 
         indices = np.argsort(perf)[perf.shape[0] - n_best:]
         return indices

autosklearn/pipeline/components/classification/decision_tree.py

@@ -32,9 +32,10 @@ def fit(self, X, y, sample_weight=None):
 
         self.max_features = float(self.max_features)
         if self.max_depth == "None":
-            self.max_depth = None
+            max_depth = self.max_depth = None
         else:
             num_features = X.shape[1]
+            self.max_depth = int(self.max_depth)
             max_depth = max(1, int(np.round(self.max_depth * num_features, 0)))
         self.min_samples_split = int(self.min_samples_split)
         self.min_samples_leaf = int(self.min_samples_leaf)

autosklearn/pipeline/components/regression/decision_tree.py

@@ -30,9 +30,10 @@ def fit(self, X, y, sample_weight=None):
 
         self.max_features = float(self.max_features)
         if self.max_depth == "None":
-            self.max_depth = None
+            max_depth = self.max_depth = None
         else:
             num_features = X.shape[1]
+            self.max_depth = int(self.max_depth)
             max_depth = max(1, int(np.round(self.max_depth * num_features, 0)))
         self.min_samples_split = int(self.min_samples_split)
         self.min_samples_leaf = int(self.min_samples_leaf)

test/test_pipeline/components/classification/test_decision_tree.py

@@ -12,22 +12,22 @@ class DecisionTreetComponentTest(unittest.TestCase):
     def test_default_configuration(self):
         for i in range(10):
             predictions, targets = _test_classifier(DecisionTree)
-            self.assertAlmostEqual(0.92,
+            self.assertAlmostEqual(0.62,
                                    sklearn.metrics.accuracy_score(predictions,
                                                                   targets))
 
     def test_default_configuration_sparse(self):
         for i in range(10):
             predictions, targets = _test_classifier(DecisionTree, sparse=True)
-            self.assertAlmostEqual(0.69999999999999996,
+            self.assertAlmostEqual(0.41999999999999998,
                                    sklearn.metrics.accuracy_score(predictions,
                                                               targets))
 
     def test_default_configuration_predict_proba(self):
         for i in range(10):
             predictions, targets = _test_classifier_predict_proba(
                 DecisionTree)
-            self.assertAlmostEqual(0.28069887755912964,
+            self.assertAlmostEqual(0.51333963481747835,
                 sklearn.metrics.log_loss(targets, predictions))
 
     def test_default_configuration_binary(self):
@@ -43,7 +43,7 @@ def test_default_configuration_multilabel(self):
             predictions, targets = _test_classifier(
                 DecisionTree, make_multilabel=True)
             print(predictions, targets)
-            self.assertAlmostEqual(0.94120857699805072,
+            self.assertAlmostEqual(0.81108108108108112,
                                    sklearn.metrics.average_precision_score(
                                        targets, predictions))
 
@@ -52,7 +52,7 @@ def test_default_configuration_multilabel_predict_proba(self):
             predictions, targets = _test_classifier_predict_proba(
                 DecisionTree, make_multilabel=True)
             self.assertEqual(predictions.shape, ((50, 3)))
-            self.assertAlmostEqual(0.94589326168273546,
+            self.assertAlmostEqual(0.83333333333333337,
                                    sklearn.metrics.average_precision_score(
                                        targets, predictions))
 

test/test_pipeline/components/data_preprocessing/test_balancing.py

@@ -68,7 +68,7 @@ def test_weighting_effect(self):
 
         for name, clf, acc_no_weighting, acc_weighting in \
                 [('adaboost', AdaboostClassifier, 0.709, 0.658),
-                 ('decision_tree', DecisionTree, 0.683, 0.701),
+                 ('decision_tree', DecisionTree, 0.724, 0.692),
                  ('extra_trees', ExtraTreesClassifier, 0.812, 0.8),
                  ('gradient_boosting', GradientBoostingClassifier,
                     0.800, 0.760),

test/test_pipeline/components/regression/test_decision_tree.py

@@ -10,13 +10,13 @@ class DecisionTreetComponentTest(unittest.TestCase):
     def test_default_configuration(self):
         for i in range(10):
             predictions, targets = _test_regressor(DecisionTree,)
-            self.assertAlmostEqual(0.14886750572325669,
+            self.assertAlmostEqual(0.1564592449511697,
                                    sklearn.metrics.r2_score(targets,
                                                             predictions))
 
     def test_default_configuration_sparse(self):
         for i in range(10):
             predictions, targets = _test_regressor(DecisionTree, sparse=True)
-            self.assertAlmostEqual(0.021778487309118133,
+            self.assertAlmostEqual(-0.020818312539637507,
                                    sklearn.metrics.r2_score(targets,
                                                             predictions))