Adress scikit-learn-contrib#176 - Fix "fit then sample" bug in pipeli…

…ne (scikit-learn-contrib#181)

imblearn/pipeline.py

@@ -230,10 +230,14 @@ def sample(self, X, y):
         Xt = X
         for _, transform in self.steps[:-1]:
             if hasattr(transform, "fit_sample"):
-                pass
+                # XXX: Calling sample in pipeline it means that the
+                # last estimator is a sampler. Samplers don't carry 
+                # the sampled data. So, call 'fit_sample' in all intermediate
+                # steps to get the sampled data for the last estimator.
+                Xt, y = transform.fit_sample(Xt, y)
             else:
                 Xt = transform.transform(Xt)
-        return self.steps[-1][-1].sample(Xt, y)
+        return self.steps[-1][-1].fit_sample(Xt, y)
 
     @if_delegate_has_method(delegate='_final_estimator')
     def predict(self, X):

imblearn/tests/test_pipeline.py

@@ -17,7 +17,7 @@
                                    assert_true, assert_warns_message)
 
 from imblearn.pipeline import Pipeline, make_pipeline
-from imblearn.under_sampling import RandomUnderSampler
+from imblearn.under_sampling import RandomUnderSampler, EditedNearestNeighbours as ENN
 
 JUNK_FOOD_DOCS = (
     "the pizza pizza beer copyright",
@@ -473,4 +473,36 @@ def test_pipeline_with_step_that_it_is_pipeline():
     filter1 = SelectKBest(f_classif, k=2)
     pipe1 = Pipeline([('rus', rus), ('anova', filter1)])
     assert_raises(TypeError, Pipeline, [('pipe1', pipe1), ('logistic', clf)])
-
+
+def test_pipeline_fit_then_sample_with_sampler_last_estimator():
+    X, y = make_classification(n_classes=2, class_sep=2, weights=[0.1, 0.9],
+        n_informative=3, n_redundant=1, flip_y=0,
+        n_features=20, n_clusters_per_class=1,
+        n_samples=50000, random_state=0)
+
+    rus = RandomUnderSampler(random_state=42)
+    enn = ENN()
+    pipeline = make_pipeline(rus, enn)
+    X_fit_sample_resampled, y_fit_sample_resampled = pipeline.fit_sample(X,y)
+    pipeline = make_pipeline(rus, enn)
+    pipeline.fit(X,y)
+    X_fit_then_sample_resampled, y_fit_then_sample_resampled = pipeline.sample(X,y)
+    assert_array_equal(X_fit_sample_resampled, X_fit_then_sample_resampled)
+    assert_array_equal(y_fit_sample_resampled, y_fit_then_sample_resampled)
+
+
+def test_pipeline_fit_then_sample_of_three_samplers_with_sampler_last_estimator():
+    X, y = make_classification(n_classes=2, class_sep=2, weights=[0.1, 0.9],
+        n_informative=3, n_redundant=1, flip_y=0,
+        n_features=20, n_clusters_per_class=1,
+        n_samples=50000, random_state=0)
+
+    rus = RandomUnderSampler(random_state=42)
+    enn = ENN()
+    pipeline = make_pipeline(rus, enn, rus)
+    X_fit_sample_resampled, y_fit_sample_resampled = pipeline.fit_sample(X,y)
+    pipeline = make_pipeline(rus, enn, rus)
+    pipeline.fit(X,y)
+    X_fit_then_sample_resampled, y_fit_then_sample_resampled = pipeline.sample(X,y)
+    assert_array_equal(X_fit_sample_resampled, X_fit_then_sample_resampled)
+    assert_array_equal(y_fit_sample_resampled, y_fit_then_sample_resampled)