chore: Compatibility scikit-learn 1.9.0 (#1169)

Author: glemaitre

imblearn/ensemble/_forest.py

@@ -586,7 +586,12 @@ def fit(self, X, y, sample_weight=None):
 
         self.n_outputs_ = y.shape[1]
 
-        y_encoded, expanded_class_weight = self._validate_y_class_weight(y)
+        if sklearn_version >= parse_version("1.9"):
+            y_encoded, expanded_class_weight = self._validate_y_class_weight(
+                y, sample_weight
+            )
+        else:
+            y_encoded, expanded_class_weight = self._validate_y_class_weight(y)
 
         if getattr(y, "dtype", None) != DOUBLE or not y.flags.contiguous:
             y_encoded = np.ascontiguousarray(y_encoded, dtype=DOUBLE)
@@ -610,9 +615,16 @@ def fit(self, X, y, sample_weight=None):
                 sample_weight = expanded_class_weight
 
         # Get bootstrap sample size
-        n_samples_bootstrap = _get_n_samples_bootstrap(
-            n_samples=X.shape[0], max_samples=self.max_samples
-        )
+        if sklearn_version >= parse_version("1.9"):
+            n_samples_bootstrap = _get_n_samples_bootstrap(
+                n_samples=X.shape[0],
+                max_samples=self.max_samples,
+                sample_weight=sample_weight,
+            )
+        else:
+            n_samples_bootstrap = _get_n_samples_bootstrap(
+                n_samples=X.shape[0], max_samples=self.max_samples
+            )
 
         # Check parameters
         self._validate_estimator()
@@ -781,13 +793,26 @@ def _compute_oob_predictions(self, X, y):
             y_resample = y[sampler.sample_indices_]
 
             n_sample_subset = y_resample.shape[0]
-            n_samples_bootstrap = _get_n_samples_bootstrap(
-                n_sample_subset, self.max_samples
-            )
+            if sklearn_version >= parse_version("1.9"):
+                n_samples_bootstrap = _get_n_samples_bootstrap(
+                    n_sample_subset, self.max_samples, sample_weight=None
+                )
+            else:
+                n_samples_bootstrap = _get_n_samples_bootstrap(
+                    n_sample_subset, self.max_samples
+                )
 
-            unsampled_indices = _generate_unsampled_indices(
-                estimator.random_state, n_sample_subset, n_samples_bootstrap
-            )
+            if sklearn_version >= parse_version("1.9"):
+                unsampled_indices = _generate_unsampled_indices(
+                    estimator.random_state,
+                    n_sample_subset,
+                    n_samples_bootstrap,
+                    sample_weight=None,
+                )
+            else:
+                unsampled_indices = _generate_unsampled_indices(
+                    estimator.random_state, n_sample_subset, n_samples_bootstrap
+                )
 
             y_pred = self._get_oob_predictions(
                 estimator, X_resample[unsampled_indices, :]

imblearn/metrics/tests/test_classification.py

@@ -7,7 +7,8 @@
 
 import numpy as np
 import pytest
-from sklearn import datasets, svm
+from sklearn import datasets
+from sklearn.linear_model import LogisticRegression
 from sklearn.metrics import (
     accuracy_score,
     average_precision_score,
@@ -43,7 +44,7 @@
 
 
 def make_prediction(dataset=None, binary=False):
-    """Make some classification predictions on a toy dataset using a SVC
+    """Make some classification predictions on a toy dataset using a classifier.
     If binary is True restrict to a binary classification problem instead of a
     multiclass classification problem
     """
@@ -72,7 +73,7 @@ def make_prediction(dataset=None, binary=False):
     X = np.c_[X, rng.randn(n_samples, 200 * n_features)]
 
     # run classifier, get class probabilities and label predictions
-    clf = svm.SVC(kernel="linear", probability=True, random_state=0)
+    clf = LogisticRegression()
     probas_pred = clf.fit(X[:half], y[:half]).predict_proba(X[half:])
 
     if binary:
@@ -275,10 +276,10 @@ def test_geometric_mean_sample_weight(
 @pytest.mark.parametrize(
     "average, expected_gmean",
     [
-        ("multiclass", 0.41),
-        (None, [0.85, 0.29, 0.7]),
-        ("macro", 0.68),
-        ("weighted", 0.65),
+        ("multiclass", 0.36),
+        (None, [0.82, 0.24, 0.72]),
+        ("macro", 0.67),
+        ("weighted", 0.64),
     ],
 )
 def test_geometric_mean_score_prediction(average, expected_gmean):
@@ -309,10 +310,10 @@ def test_classification_report_imbalanced_multiclass():
 
     # print classification report with class names
     expected_report = (
-        "pre rec spe f1 geo iba sup setosa 0.83 0.79 0.92 "
-        "0.81 0.85 0.72 24 versicolor 0.33 0.10 0.86 0.15 "
-        "0.29 0.08 31 virginica 0.42 0.90 0.55 0.57 0.70 "
-        "0.51 20 avg / total 0.51 0.53 0.80 0.47 0.58 0.40 75"
+        "pre rec spe f1 geo iba sup setosa 0.70 0.79 0.84 "
+        "0.75 0.82 0.66 24 versicolor 0.29 0.06 0.89 0.11 "
+        "0.24 0.05 31 virginica 0.44 0.90 0.58 0.59 0.72 "
+        "0.54 20 avg / total 0.46 0.52 0.79 0.44 0.55 0.38 75"
     )
 
     report = classification_report_imbalanced(
@@ -324,10 +325,10 @@ def test_classification_report_imbalanced_multiclass():
     assert _format_report(report) == expected_report
     # print classification report with label detection
     expected_report = (
-        "pre rec spe f1 geo iba sup 0 0.83 0.79 0.92 0.81 "
-        "0.85 0.72 24 1 0.33 0.10 0.86 0.15 0.29 0.08 31 "
-        "2 0.42 0.90 0.55 0.57 0.70 0.51 20 avg / total "
-        "0.51 0.53 0.80 0.47 0.58 0.40 75"
+        "pre rec spe f1 geo iba sup 0 0.70 0.79 0.84 0.75 "
+        "0.82 0.66 24 1 0.29 0.06 0.89 0.11 0.24 0.05 31 "
+        "2 0.44 0.90 0.58 0.59 0.72 0.54 20 avg / total "
+        "0.46 0.52 0.79 0.44 0.55 0.38 75"
     )
 
     report = classification_report_imbalanced(y_true, y_pred)
@@ -340,12 +341,12 @@ def test_classification_report_imbalanced_multiclass_with_digits():
 
     # print classification report with class names
     expected_report = (
-        "pre rec spe f1 geo iba sup setosa 0.82609 0.79167 "
-        "0.92157 0.80851 0.85415 0.72010 24 versicolor "
-        "0.33333 0.09677 0.86364 0.15000 0.28910 0.07717 "
-        "31 virginica 0.41860 0.90000 0.54545 0.57143 0.70065 "
-        "0.50831 20 avg / total 0.51375 0.53333 0.79733 "
-        "0.47310 0.57966 0.39788 75"
+        "pre rec spe f1 geo iba sup setosa 0.70370 0.79167 "
+        "0.84314 0.74510 0.81700 0.66405 24 versicolor "
+        "0.28571 0.06452 0.88636 0.10526 0.23913 0.05249 "
+        "31 virginica 0.43902 0.90000 0.58182 0.59016 0.72363 "
+        "0.54030 20 avg / total 0.46035 0.52000 0.79132 "
+        "0.43932 0.55325 0.37827 75"
     )
     report = classification_report_imbalanced(
         y_true,
@@ -357,10 +358,10 @@ def test_classification_report_imbalanced_multiclass_with_digits():
     assert _format_report(report) == expected_report
     # print classification report with label detection
     expected_report = (
-        "pre rec spe f1 geo iba sup 0 0.83 0.79 0.92 0.81 "
-        "0.85 0.72 24 1 0.33 0.10 0.86 0.15 0.29 0.08 31 "
-        "2 0.42 0.90 0.55 0.57 0.70 0.51 20 avg / total 0.51 "
-        "0.53 0.80 0.47 0.58 0.40 75"
+        "pre rec spe f1 geo iba sup 0 0.70 0.79 0.84 0.75 "
+        "0.82 0.66 24 1 0.29 0.06 0.89 0.11 0.24 0.05 31 "
+        "2 0.44 0.90 0.58 0.59 0.72 0.54 20 avg / total 0.46 "
+        "0.52 0.79 0.44 0.55 0.38 75"
     )
     report = classification_report_imbalanced(y_true, y_pred)
     assert _format_report(report) == expected_report
@@ -373,19 +374,19 @@ def test_classification_report_imbalanced_multiclass_with_string_label():
     y_pred = np.array(["blue", "green", "red"])[y_pred]
 
     expected_report = (
-        "pre rec spe f1 geo iba sup blue 0.83 0.79 0.92 0.81 "
-        "0.85 0.72 24 green 0.33 0.10 0.86 0.15 0.29 0.08 31 "
-        "red 0.42 0.90 0.55 0.57 0.70 0.51 20 avg / total "
-        "0.51 0.53 0.80 0.47 0.58 0.40 75"
+        "pre rec spe f1 geo iba sup blue 0.70 0.79 0.84 0.75 "
+        "0.82 0.66 24 green 0.29 0.06 0.89 0.11 0.24 0.05 31 "
+        "red 0.44 0.90 0.58 0.59 0.72 0.54 20 avg / total "
+        "0.46 0.52 0.79 0.44 0.55 0.38 75"
     )
     report = classification_report_imbalanced(y_true, y_pred)
     assert _format_report(report) == expected_report
 
     expected_report = (
-        "pre rec spe f1 geo iba sup a 0.83 0.79 0.92 0.81 0.85 "
-        "0.72 24 b 0.33 0.10 0.86 0.15 0.29 0.08 31 c 0.42 "
-        "0.90 0.55 0.57 0.70 0.51 20 avg / total 0.51 0.53 "
-        "0.80 0.47 0.58 0.40 75"
+        "pre rec spe f1 geo iba sup a 0.70 0.79 0.84 0.75 0.82 "
+        "0.66 24 b 0.29 0.06 0.89 0.11 0.24 0.05 31 c 0.44 "
+        "0.90 0.58 0.59 0.72 0.54 20 avg / total 0.46 0.52 "
+        "0.79 0.44 0.55 0.38 75"
     )
     report = classification_report_imbalanced(
         y_true, y_pred, target_names=["a", "b", "c"]
@@ -401,10 +402,10 @@ def test_classification_report_imbalanced_multiclass_with_unicode_label():
     y_pred = labels[y_pred]
 
     expected_report = (
-        "pre rec spe f1 geo iba sup blue¢ 0.83 0.79 0.92 0.81 "
-        "0.85 0.72 24 green¢ 0.33 0.10 0.86 0.15 0.29 0.08 31 "
-        "red¢ 0.42 0.90 0.55 0.57 0.70 0.51 20 avg / total "
-        "0.51 0.53 0.80 0.47 0.58 0.40 75"
+        "pre rec spe f1 geo iba sup blue¢ 0.70 0.79 0.84 0.75 "
+        "0.82 0.66 24 green¢ 0.29 0.06 0.89 0.11 0.24 0.05 31 "
+        "red¢ 0.44 0.90 0.58 0.59 0.72 0.54 20 avg / total "
+        "0.46 0.52 0.79 0.44 0.55 0.38 75"
     )
     report = classification_report_imbalanced(y_true, y_pred)
     assert _format_report(report) == expected_report
@@ -418,10 +419,10 @@ def test_classification_report_imbalanced_multiclass_with_long_string_label():
     y_pred = labels[y_pred]
 
     expected_report = (
-        "pre rec spe f1 geo iba sup blue 0.83 0.79 0.92 0.81 "
-        "0.85 0.72 24 greengreengreengreengreen 0.33 0.10 "
-        "0.86 0.15 0.29 0.08 31 red 0.42 0.90 0.55 0.57 0.70 "
-        "0.51 20 avg / total 0.51 0.53 0.80 0.47 0.58 0.40 75"
+        "pre rec spe f1 geo iba sup blue 0.70 0.79 0.84 0.75 "
+        "0.82 0.66 24 greengreengreengreengreen 0.29 0.06 "
+        "0.89 0.11 0.24 0.05 31 red 0.44 0.90 0.58 0.59 0.72 "
+        "0.54 20 avg / total 0.46 0.52 0.79 0.44 0.55 0.38 75"
     )
 
     report = classification_report_imbalanced(y_true, y_pred)

imblearn/over_sampling/tests/test_random_over_sampler.py

@@ -308,4 +308,4 @@ def test_random_over_sampler_full_nat():
     assert X_res.shape == (4, 2)
     assert y_res.shape == (4,)
 
-    assert X_res["col_timedelta"].dtype == "timedelta64[ns]"
+    assert X_res["col_timedelta"].dtype.kind == "m"  # timedelta

imblearn/tests/test_pipeline.py

@@ -350,10 +350,10 @@ def test_pipeline_methods_pca_svm():
     iris = load_iris()
     X = iris.data
     y = iris.target
-    # Test with PCA + SVC
-    clf = SVC(gamma="scale", probability=True, random_state=0)
+    # Test with PCA + LogisticRegression
+    clf = LogisticRegression()
     pca = PCA(svd_solver="full", n_components="mle", whiten=True)
-    pipe = Pipeline([("pca", pca), ("svc", clf)])
+    pipe = Pipeline([("pca", pca), ("clf", clf)])
     pipe.fit(X, y)
     pipe.predict(X)
     pipe.predict_proba(X)
@@ -370,15 +370,10 @@ def test_pipeline_methods_preprocessing_svm():
     n_classes = len(np.unique(y))
     scaler = StandardScaler()
     pca = PCA(n_components=2, svd_solver="randomized", whiten=True)
-    clf = SVC(
-        gamma="scale",
-        probability=True,
-        random_state=0,
-        decision_function_shape="ovr",
-    )
+    clf = LogisticRegression()
 
     for preprocessing in [scaler, pca]:
-        pipe = Pipeline([("preprocess", preprocessing), ("svc", clf)])
+        pipe = Pipeline([("preprocess", preprocessing), ("clf", clf)])
         pipe.fit(X, y)
 
         # check shapes of various prediction functions
@@ -671,11 +666,11 @@ def test_pipeline_memory_transformer():
     cachedir = mkdtemp()
     try:
         memory = Memory(cachedir, verbose=10)
-        # Test with Transformer + SVC
-        clf = SVC(gamma="scale", probability=True, random_state=0)
+        # Test with Transformer + LogisticRegression
+        clf = LogisticRegression()
         transf = DummyTransf()
-        pipe = Pipeline([("transf", clone(transf)), ("svc", clf)])
-        cached_pipe = Pipeline([("transf", transf), ("svc", clf)], memory=memory)
+        pipe = Pipeline([("transf", clone(transf)), ("clf", clf)])
+        cached_pipe = Pipeline([("transf", transf), ("clf", clf)], memory=memory)
 
         # Memoize the transformer at the first fit
         cached_pipe.fit(X, y)
@@ -707,10 +702,10 @@ def test_pipeline_memory_transformer():
         assert cached_pipe.named_steps["transf"].timestamp_ == expected_ts
         # Create a new pipeline with cloned estimators
         # Check that even changing the name step does not affect the cache hit
-        clf_2 = SVC(gamma="scale", probability=True, random_state=0)
+        clf_2 = LogisticRegression(random_state=0)
         transf_2 = DummyTransf()
         cached_pipe_2 = Pipeline(
-            [("transf_2", transf_2), ("svc", clf_2)], memory=memory
+            [("transf_2", transf_2), ("clf", clf_2)], memory=memory
         )
         cached_pipe_2.fit(X, y)
 
@@ -746,11 +741,11 @@ def test_pipeline_memory_sampler():
     cachedir = mkdtemp()
     try:
         memory = Memory(cachedir, verbose=10)
-        # Test with Transformer + SVC
-        clf = SVC(gamma="scale", probability=True, random_state=0)
+        # Test with Sampler + LogisticRegression
+        clf = LogisticRegression()
         transf = DummySampler()
-        pipe = Pipeline([("transf", clone(transf)), ("svc", clf)])
-        cached_pipe = Pipeline([("transf", transf), ("svc", clf)], memory=memory)
+        pipe = Pipeline([("transf", clone(transf)), ("clf", clf)])
+        cached_pipe = Pipeline([("transf", transf), ("clf", clf)], memory=memory)
 
         # Memoize the transformer at the first fit
         cached_pipe.fit(X, y)
@@ -782,10 +777,10 @@ def test_pipeline_memory_sampler():
         assert cached_pipe.named_steps["transf"].timestamp_ == expected_ts
         # Create a new pipeline with cloned estimators
         # Check that even changing the name step does not affect the cache hit
-        clf_2 = SVC(gamma="scale", probability=True, random_state=0)
+        clf_2 = LogisticRegression(random_state=0)
         transf_2 = DummySampler()
         cached_pipe_2 = Pipeline(
-            [("transf_2", transf_2), ("svc", clf_2)], memory=memory
+            [("transf_2", transf_2), ("clf", clf_2)], memory=memory
         )
         cached_pipe_2.fit(X, y)
 
@@ -820,11 +815,11 @@ def test_pipeline_methods_pca_rus_svm():
         random_state=0,
     )
 
-    # Test with PCA + SVC
-    clf = SVC(gamma="scale", probability=True, random_state=0)
+    # Test with PCA + LogisticRegression
+    clf = LogisticRegression()
     pca = PCA()
     rus = RandomUnderSampler(random_state=0)
-    pipe = Pipeline([("pca", pca), ("rus", rus), ("svc", clf)])
+    pipe = Pipeline([("pca", pca), ("rus", rus), ("clf", clf)])
     pipe.fit(X, y)
     pipe.predict(X)
     pipe.predict_proba(X)
@@ -847,11 +842,11 @@ def test_pipeline_methods_rus_pca_svm():
         random_state=0,
     )
 
-    # Test with PCA + SVC
-    clf = SVC(gamma="scale", probability=True, random_state=0)
+    # Test with PCA + LogisticRegression
+    clf = LogisticRegression()
     pca = PCA()
     rus = RandomUnderSampler(random_state=0)
-    pipe = Pipeline([("rus", rus), ("pca", pca), ("svc", clf)])
+    pipe = Pipeline([("rus", rus), ("pca", pca), ("clf", clf)])
     pipe.fit(X, y)
     pipe.predict(X)
     pipe.predict_proba(X)

imblearn/under_sampling/_prototype_selection/tests/test_random_under_sampler.py

@@ -183,4 +183,4 @@ def test_random_under_sampler_full_nat():
     assert X_res.shape == (2, 2)
     assert y_res.shape == (2,)
 
-    assert X_res["col_timedelta"].dtype == "timedelta64[ns]"
+    assert X_res["col_timedelta"].dtype.kind == "m"  # timedelta

imblearn/utils/_validation.py

@@ -78,8 +78,8 @@ def _transfrom_one(self, array, props):
                 for col in ret.columns:
                     if (
                         ret[col].isnull().all()
-                        and ret[col].dtype == "datetime64[ns]"
-                        and props["dtypes"][col] == "timedelta64[ns]"
+                        and ret[col].dtype.kind == "M"  # datetime64
+                        and props["dtypes"][col].kind == "m"  # timedelta64
                     ):
                         ret[col] = pd.to_timedelta(["NaT"] * len(ret[col]))
                 # try again