Merge pull request #766 from TwsThomas/patch-1

pass skipna=False explicitly

Author: mfeurer

openml/datasets/functions.py

@@ -528,7 +528,7 @@ def attributes_arff_from_df(df):
     for column_name in df:
         # skipna=True does not infer properly the dtype. The NA values are
         # dropped before the inference instead.
-        column_dtype = pd.api.types.infer_dtype(df[column_name].dropna())
+        column_dtype = pd.api.types.infer_dtype(df[column_name].dropna(), skipna=False)
 
         if column_dtype == 'categorical':
             # for categorical feature, arff expects a list string. However, a