FIX reproducibility and parallelization of InstanceHardnessThreshold

doc/whats_new/v0.6.rst

@@ -15,6 +15,14 @@ scikit-learn:
 - :class:`imblearn.under_sampling.ClusterCentroids`
 - :class:`imblearn.under_sampling.InstanceHardnessThreshold`
 
+Bug fixes
+.........
+
+- :class:`imblearn.under_sampling.InstanceHardnessThreshold` now take into
+  account the `random_state` and will give deterministic results. In addition,
+  `cross_val_predict` is used to take advantage of the parallelism.
+  :pr:`599` by :user:`Shihab Shahriar Khan <Shihab-Shahriar>`.
+
 Maintenance
 ...........
 

imblearn/under_sampling/_prototype_selection/_instance_hardness_threshold.py

@@ -12,7 +12,10 @@
 
 from sklearn.base import ClassifierMixin, clone
 from sklearn.ensemble import RandomForestClassifier
+from sklearn.ensemble._base import _set_random_states
 from sklearn.model_selection import StratifiedKFold
+from sklearn.model_selection import cross_val_predict
+from sklearn.utils import check_random_state
 from sklearn.utils import _safe_indexing
 
 from ..base import BaseUnderSampler
@@ -108,7 +111,7 @@ def __init__(
         self.cv = cv
         self.n_jobs = n_jobs
 
-    def _validate_estimator(self):
+    def _validate_estimator(self, random_state):
         """Private function to create the classifier"""
 
         if (
@@ -117,6 +120,8 @@ def _validate_estimator(self):
             and hasattr(self.estimator, "predict_proba")
         ):
             self.estimator_ = clone(self.estimator)
+            _set_random_states(self.estimator_, random_state)
+
         elif self.estimator is None:
             self.estimator_ = RandomForestClassifier(
                 n_estimators=100,
@@ -131,22 +136,18 @@ def _validate_estimator(self):
             )
 
     def _fit_resample(self, X, y):
-        self._validate_estimator()
+        random_state = check_random_state(self.random_state)
+        self._validate_estimator(random_state)
 
         target_stats = Counter(y)
-        skf = StratifiedKFold(n_splits=self.cv, shuffle=False).split(X, y)
-        probabilities = np.zeros(y.shape[0], dtype=float)
-
-        for train_index, test_index in skf:
-            X_train = _safe_indexing(X, train_index)
-            X_test = _safe_indexing(X, test_index)
-            y_train = _safe_indexing(y, train_index)
-            y_test = _safe_indexing(y, test_index)
-
-            self.estimator_.fit(X_train, y_train)
-
-            probs = self.estimator_.predict_proba(X_test)
-            probabilities[test_index] = probs[range(len(y_test)), y_test]
+        skf = StratifiedKFold(
+            n_splits=self.cv, shuffle=True, random_state=random_state,
+        )
+        probabilities = cross_val_predict(
+            self.estimator_, X, y, cv=skf, n_jobs=self.n_jobs,
+            method='predict_proba'
+        )
+        probabilities = probabilities[range(len(y)), y]
 
         idx_under = np.empty((0,), dtype=int)
 

imblearn/under_sampling/_prototype_selection/tests/test_instance_hardness_threshold.py

@@ -6,7 +6,10 @@
 import pytest
 import numpy as np
 
+
 from sklearn.ensemble import GradientBoostingClassifier
+from sklearn.ensemble import RandomForestClassifier
+from sklearn.utils._testing import assert_array_equal
 
 from imblearn.under_sampling import InstanceHardnessThreshold
 
@@ -76,3 +79,16 @@ def test_iht_fit_resample_wrong_class_obj():
     iht = InstanceHardnessThreshold(estimator=est, random_state=RND_SEED)
     with pytest.raises(ValueError, match="Invalid parameter `estimator`"):
         iht.fit_resample(X, Y)
+
+
+def test_iht_reproducibility():
+    from sklearn.datasets import load_digits
+    X_digits, y_digits = load_digits(return_X_y=True)
+    idx_sampled = []
+    for seed in range(5):
+        est = RandomForestClassifier(n_estimators=10, random_state=seed)
+        iht = InstanceHardnessThreshold(estimator=est, random_state=RND_SEED)
+        iht.fit_resample(X_digits, y_digits)
+        idx_sampled.append(iht.sample_indices_.copy())
+    for idx_1, idx_2 in zip(idx_sampled, idx_sampled[1:]):
+        assert_array_equal(idx_1, idx_2)