EHN allow to become minority in AllKNN (#313)

Author: glemaitre

imblearn/under_sampling/prototype_selection/edited_nearest_neighbours.py

@@ -477,6 +477,12 @@ class AllKNN(BaseCleaningSampler):
         - If ``'mode'``, the majority vote of the neighbours will be used in
           order to exclude a sample.
 
+    allow_minority : bool, optional (default=False)
+        If ``True``, it allows the majority classes to become the minority
+        class without early stopping.
+
+        .. versionadded:: 0.3
+
     n_jobs : int, optional (default=-1)
         The number of thread to open when it is possible.
 
@@ -519,12 +525,14 @@ def __init__(self,
                  size_ngh=None,
                  n_neighbors=3,
                  kind_sel='all',
+                 allow_minority=False,
                  n_jobs=-1):
         super(AllKNN, self).__init__(ratio=ratio, random_state=random_state)
         self.return_indices = return_indices
         self.size_ngh = size_ngh
         self.n_neighbors = n_neighbors
         self.kind_sel = kind_sel
+        self.allow_minority = allow_minority
         self.n_jobs = n_jobs
 
     def _validate_estimator(self):
@@ -595,6 +603,9 @@ def _sample(self, X, y):
             ])
             b_min_bec_maj = np.any(count_non_min <
                                    target_stats[class_minority])
+            if self.allow_minority:
+                # overwrite b_min_bec_maj
+                b_min_bec_maj = False
 
             # Case 2
             b_remove_maj_class = (len(stats_enn) < len(target_stats))

imblearn/under_sampling/prototype_selection/tests/test_allknn.py

@@ -7,8 +7,9 @@
 
 import numpy as np
 from sklearn.utils.testing import (assert_allclose, assert_array_equal,
-                                   assert_raises)
+                                   assert_raises, assert_true)
 from sklearn.neighbors import NearestNeighbors
+from sklearn.datasets import make_classification
 
 from imblearn.under_sampling import AllKNN
 
@@ -66,6 +67,19 @@ def test_allknn_fit_sample():
     assert_allclose(y_resampled, y_gt, rtol=R_TOL)
 
 
+def test_all_knn_allow_minority():
+    X, y = make_classification(n_samples=10000, n_features=2, n_informative=2,
+                               n_redundant=0, n_repeated=0, n_classes=3,
+                               n_clusters_per_class=1, weights=[0.2, 0.3, 0.5],
+                               class_sep=0.4, random_state=0)
+
+    allknn = AllKNN(random_state=RND_SEED, allow_minority=True)
+    X_res_1, y_res_1 = allknn.fit_sample(X, y)
+    allknn = AllKNN(random_state=RND_SEED)
+    X_res_2, y_res_2 = allknn.fit_sample(X, y)
+    assert_true(len(y_res_1) < len(y_res_2))
+
+
 def test_allknn_fit_sample_with_indices():
     allknn = AllKNN(return_indices=True, random_state=RND_SEED)
     X_resampled, y_resampled, idx_under = allknn.fit_sample(X, Y)