FIX handle sparse matrices in SMOTEN (#1003)

Author: glemaitre

doc/whats_new/v0.11.rst

@@ -13,6 +13,10 @@ Bug fixes
   `bool` and `pd.category` by delegating the conversion to scikit-learn encoder.
   :pr:`1002` by :user:`Guillaume Lemaitre <glemaitre>`.
 
+- Handle sparse matrices in :class:`~imblearn.over_sampling.SMOTEN` and raise a warning
+  since it requires a conversion to dense matrices.
+  :pr:`1003` by :user:`Guillaume Lemaitre <glemaitre>`.
+
 Compatibility
 .............
 

imblearn/over_sampling/_smote/base.py

@@ -14,6 +14,7 @@
 import numpy as np
 from scipy import sparse
 from sklearn.base import clone
+from sklearn.exceptions import DataConversionWarning
 from sklearn.preprocessing import OneHotEncoder, OrdinalEncoder
 from sklearn.utils import _safe_indexing, check_array, check_random_state
 from sklearn.utils.sparsefuncs_fast import (
@@ -893,7 +894,7 @@ def _check_X_y(self, X, y):
             y,
             reset=True,
             dtype=None,
-            accept_sparse=False,
+            accept_sparse=["csr", "csc"],
         )
         return X, y, binarize_y
 
@@ -927,6 +928,17 @@ def _fit_resample(self, X, y):
                 FutureWarning,
             )
 
+        if sparse.issparse(X):
+            X_sparse_format = X.format
+            X = X.toarray()
+            warnings.warn(
+                "Passing a sparse matrix to SMOTEN is not really efficient since it is"
+                " converted to a dense array internally.",
+                DataConversionWarning,
+            )
+        else:
+            X_sparse_format = None
+
         self._validate_estimator()
 
         X_resampled = [X.copy()]
@@ -964,7 +976,12 @@ def _fit_resample(self, X, y):
         X_resampled = np.vstack(X_resampled)
         y_resampled = np.hstack(y_resampled)
 
-        return X_resampled, y_resampled
+        if X_sparse_format == "csr":
+            return sparse.csr_matrix(X_resampled), y_resampled
+        elif X_sparse_format == "csc":
+            return sparse.csc_matrix(X_resampled), y_resampled
+        else:
+            return X_resampled, y_resampled
 
     def _more_tags(self):
         return {"X_types": ["2darray", "dataframe", "string"]}

imblearn/over_sampling/_smote/tests/test_smoten.py

@@ -1,6 +1,8 @@
 import numpy as np
 import pytest
-from sklearn.preprocessing import OrdinalEncoder
+from sklearn.exceptions import DataConversionWarning
+from sklearn.preprocessing import OneHotEncoder, OrdinalEncoder
+from sklearn.utils._testing import _convert_container
 
 from imblearn.over_sampling import SMOTEN
 
@@ -56,6 +58,24 @@ def test_smoten_resampling():
     np.testing.assert_array_equal(y_generated, "not apple")
 
 
+@pytest.mark.parametrize("sparse_format", ["sparse_csr", "sparse_csc"])
+def test_smoten_sparse_input(data, sparse_format):
+    """Check that we handle sparse input in SMOTEN even if it is not efficient.
+
+    Non-regression test for:
+    https://github.com/scikit-learn-contrib/imbalanced-learn/issues/971
+    """
+    X, y = data
+    X = OneHotEncoder().fit_transform(X)
+    X = _convert_container(X, sparse_format)
+
+    with pytest.warns(DataConversionWarning, match="is not really efficient"):
+        X_res, y_res = SMOTEN(random_state=0).fit_resample(X, y)
+
+    assert X_res.format == X.format
+    assert X_res.shape[0] == len(y_res)
+
+
 def test_smoten_categorical_encoder(data):
     """Check that `categorical_encoder` is used when provided."""