Convert non-str column names to str when creating a dataset. (#851)

* Convert non-str column names to str when creating a dataset.

* Add unit test

Author: PGijsbers

openml/datasets/functions.py

@@ -549,6 +549,11 @@ def attributes_arff_from_df(df):
         'string': 'STRING'
     }
     attributes_arff = []
+
+    if not all([isinstance(column_name, str) for column_name in df.columns]):
+        logger.warning("Converting non-str column names to str.")
+        df.columns = [str(column_name) for column_name in df.columns]
+
     for column_name in df:
         # skipna=True does not infer properly the dtype. The NA values are
         # dropped before the inference instead.

tests/test_datasets/test_dataset_functions.py

@@ -569,6 +569,12 @@ def test_attributes_arff_from_df(self):
         self.assertEqual(attributes, [('integer', 'INTEGER'),
                                       ('floating', 'REAL')])
 
+    def test_attributes_arff_from_df_numeric_column(self):
+        # Test column names are automatically converted to str if needed (#819)
+        df = pd.DataFrame({0: [1, 2, 3], 0.5: [4, 5, 6], 'target': [0, 1, 1]})
+        attributes = attributes_arff_from_df(df)
+        self.assertEqual(attributes, [('0', 'INTEGER'), ('0.5', 'INTEGER'), ('target', 'INTEGER')])
+
     def test_attributes_arff_from_df_mixed_dtype_categories(self):
         # liac-arff imposed categorical attributes to be of sting dtype. We
         # raise an error if this is not the case.