edit _transform_count_encode

category_encoders/count.py

@@ -271,9 +271,11 @@ def _fit_count_encode(self, X_in, y):
     def _transform_count_encode(self, X_in, y):
         """Perform the transform count encoding."""
         X = X_in.copy(deep=True)
-        X.fillna(value=np.nan, inplace=True)
 
         for col in self.cols:
+
+            X[col] = X.fillna(value=np.nan)[col]
+
             if self._min_group_size is not None:
                 if col in self._min_group_categories.keys():
                     X[col] = (

tests/test_encoders.py

@@ -692,3 +692,15 @@ def test_columns(self):
         self.assertTrue(result['float_edge'].min() < 1, 'should still be a number and untouched')
         self.assertTrue(result['unique_int'].min() < 1, 'should still be a number and untouched')
         self.assertTrue(result[321].min() < 1, 'should still be a number')
+
+    def test_ignored_columns_are_untouched(self):
+        # Make sure None values in ignored columns are preserved.
+        # See: https://github.com/scikit-learn-contrib/category_encoders/pull/261
+        X = pd.DataFrame({'col1': ['A', 'B', None], 'col2': ['C', 'D', None]})
+        y = [1, 0, 1]
+
+        for encoder_name in (set(encoders.__all__)):
+            with self.subTest(encoder_name=encoder_name):
+                enc = getattr(encoders, encoder_name)(cols=['col1'])
+                out = enc.fit_transform(X, y)
+                self.assertTrue(out.col2[2] is None)