Add a test to trigger the bug

Signed-off-by: Pierre Bartet <[email protected]>

tests/test_sklearn_pipeline.py

@@ -1335,6 +1335,32 @@ def test_pipeline_make_column_selector(self):
         )
         assert_almost_equal(expected, got[0])
 
+    @unittest.skipIf(TARGET_OPSET < 11, reason="SequenceConstruct not available")
+    @unittest.skipIf(not check_scikit_version(), reason="Scikit 0.21 too old")
+    @ignore_warnings(category=(FutureWarning, UserWarning))
+    def test_pipeline_empty_make_column_selector(self):
+        X = pandas.DataFrame(
+            {"city": ["London", "London", "Paris", "Sallisaw"]}
+        )
+
+        ct = make_column_transformer(
+            (StandardScaler(), make_column_selector(dtype_include=numpy.number)),
+            (OneHotEncoder(), make_column_selector(dtype_include=object)),
+        )
+        expected = ct.fit_transform(X)
+        onx = to_onnx(ct, X, target_opset=TARGET_OPSET)
+        sess = InferenceSession(
+            onx.SerializeToString(), providers=["CPUExecutionProvider"]
+        )
+        names = [i.name for i in sess.get_inputs()]
+        got = sess.run(
+            None,
+            {
+                names[0]: X[names[0]].values.reshape((-1, 1)),
+            },
+        )
+        assert_almost_equal(expected, got[0])
+
     @unittest.skipIf(not check_scikit_version(), reason="Scikit 0.21 too old")
     def test_feature_selector_no_converter(self):
         class ColumnSelector(TransformerMixin, BaseEstimator):