Merge pull request bsc-wdc#350 from bsc-wdc/rf_score_synchronization

Synchronizing the score in RF

dislib/classification/csvm/base.py

@@ -192,6 +192,8 @@ def score(self, x, y, collect=False):
             Test samples.
         y : ds-array, shape=(n_samples, 1)
             True labels for x.
+        collect : bool
+            When True, a synchronized result is returned.
 
         Returns
         -------

dislib/classification/rf/forest.py

@@ -197,7 +197,7 @@ def predict(self, x):
 
         return y_pred
 
-    def score(self, x, y):
+    def score(self, x, y, collect=False):
         """Accuracy classification score.
 
         Returns the mean accuracy on the given test data.
@@ -209,6 +209,8 @@ def score(self, x, y):
             The training input samples.
         y : ds-array, shape (n_samples, 1)
             The true labels.
+        collect : bool
+            When True, a synchronized result is returned.
 
         Returns
         -------
@@ -235,7 +237,9 @@ def score(self, x, y):
                                                 *tree_predictions)
                 partial_scores.append(subset_score)
 
-        return _merge_scores(*partial_scores)
+        score = _merge_scores(*partial_scores)
+
+        return compss_wait_on(score) if collect else score
 
 
 @task(returns=1)

tests/test_rf.py

@@ -1,6 +1,7 @@
 import unittest
 
 import numpy as np
+from parameterized import parameterized
 from pycompss.api.api import compss_wait_on
 from sklearn import datasets
 from sklearn.datasets import make_classification
@@ -180,7 +181,8 @@ def test_make_classification_hard_vote_score_mix(self):
         accuracy = compss_wait_on(rf.score(x_test, y_test))
         self.assertGreater(accuracy, 0.7)
 
-    def test_iris(self):
+    @parameterized.expand([(True,), (False,)])
+    def test_score_on_iris(self, collect):
         """Tests RandomForestClassifier with a minimal example."""
         x, y = datasets.load_iris(return_X_y=True)
         ds_fit = ds.array(x[::2], block_size=(30, 2))
@@ -191,7 +193,9 @@ def test_iris(self):
         rf = RandomForestClassifier(n_estimators=1, max_depth=1,
                                     random_state=0)
         rf.fit(ds_fit, fit_y)
-        accuracy = compss_wait_on(rf.score(ds_validate, validate_y))
+        accuracy = rf.score(ds_validate, validate_y, collect)
+        if not collect:
+            accuracy = compss_wait_on(accuracy)
 
         # Accuracy should be <= 2/3 for any seed, often exactly equal.
         self.assertAlmostEqual(accuracy, 2 / 3)