Added a check for wrong column names and a test for this check (#586)

* Added a check for wrong column names and a test for this check

* Old python versions are so ugly...

* Make sure to check only once converted to strings

tests/units/utilities/test_dataframe_functions.py

@@ -138,6 +138,12 @@ def test_with_wrong_input(self):
         self.assertRaises(ValueError, dataframe_functions._normalize_input_to_internal_representation, test_df,
                           None, None, None, "value")
 
+        test_df = pd.DataFrame([{"id": 0, "a_": 3, "b": 5, "sort": 1}])
+        self.assertRaises(ValueError, dataframe_functions._normalize_input_to_internal_representation, test_df, "id", "sort", None, None)
+
+        test_df = pd.DataFrame([{"id": 0, "a__c": 3, "b": 5, "sort": 1}])
+        self.assertRaises(ValueError, dataframe_functions._normalize_input_to_internal_representation, test_df, "id", "sort", None, None)
+
         test_df = pd.DataFrame([{"id": 0}])
         self.assertRaises(ValueError, dataframe_functions._normalize_input_to_internal_representation, test_df,
                           "id", None, None, None)

tsfresh/utilities/dataframe_functions.py

@@ -354,6 +354,14 @@ def _normalize_input_to_internal_representation(timeseries_container, column_id,
     # The kind columns should always be of type "str" to make the inference of feature settings later in `from_columns`
     # work
     timeseries_container[column_kind] = timeseries_container[column_kind].astype(str)
+
+    # Make sure we have only parsable names
+    for kind in timeseries_container[column_kind].unique():
+        if kind.endswith("_"):
+            raise ValueError("The kind {kind} is not allowed to end with '_'".format(kind=kind))
+        if "__" in kind:
+            raise ValueError("The kind {kind} is not allowed to contain '__'".format(kind=kind))
+
     return timeseries_container, column_id, column_kind, column_value