[MRG] Fix ValueDifferenceMetric feature validation

doc/whats_new/v0.15.rst

@@ -11,6 +11,11 @@ Changelog
 Bug fixes
 .........
 
+- Fix :meth:`~imblearn.metrics.pairwise.ValueDifferenceMetric.pairwise` to
+  validate the number of features in `X` and `Y` instead of silently ignoring
+  extra features.
+  :issue:`1198` by :user:`A Aswanth Raj <aswanth-07>`.
+
 Enhancements
 ............
 

imblearn/metrics/pairwise.py

@@ -210,11 +210,15 @@ def pairwise(self, X, Y=None):
             The VDM pairwise distance.
         """
         check_is_fitted(self)
-        X = check_array(X, ensure_non_negative=True, dtype=np.int32)
+        X = validate_data(
+            self, X=X, reset=False, ensure_non_negative=True, dtype=np.int32
+        )
         n_samples_X = X.shape[0]
 
         if Y is not None:
-            Y = check_array(Y, ensure_non_negative=True, dtype=np.int32)
+            Y = validate_data(
+                self, X=Y, reset=False, ensure_non_negative=True, dtype=np.int32
+            )
             n_samples_Y = Y.shape[0]
         else:
             n_samples_Y = n_samples_X

imblearn/metrics/tests/test_pairwise.py

@@ -170,3 +170,15 @@ def test_value_difference_value_unfitted(data):
 
     with pytest.raises(NotFittedError):
         ValueDifferenceMetric().pairwise(X_encoded)
+
+
+@pytest.mark.parametrize("invalid_arg", ["X", "Y"])
+def test_value_difference_metric_wrong_n_features(data, invalid_arg):
+    X, y = data
+    X_encoded = OrdinalEncoder(dtype=np.int32).fit_transform(X)
+    X_invalid = np.column_stack([X_encoded, np.zeros(X_encoded.shape[0])])
+    vdm = ValueDifferenceMetric().fit(X_encoded, y)
+
+    pairwise_args = (X_invalid,) if invalid_arg == "X" else (X_encoded, X_invalid)
+    with pytest.raises(ValueError, match="X has 4 features"):
+        vdm.pairwise(*pairwise_args)