Fix #378 (#418)

* Fix ascii decoding problem with python 2

* Created separate pickle files for splits and datasets according to the python version

* Added production task to unit test

* Update test_task_functions.py

Author: ArlindKadra

openml/datasets/data_feature.py

@@ -1,3 +1,4 @@
+import six
 
 class OpenMLDataFeature(object):
     """Data Feature (a.k.a. Attribute) object.
@@ -29,7 +30,11 @@ def __init__(self, index, name, data_type, nominal_values,
             raise ValueError('number_missing_values is of wrong datatype')
 
         self.index = index
-        self.name = str(name)
+        # In case of python version lower than 3, change the default ASCII encoder.
+        if six.PY2:
+            self.name = str(name.encode('utf8'))
+        else:
+            self.name = str(name)
         self.data_type = str(data_type)
         self.nominal_values = nominal_values
         self.number_missing_values = number_missing_values

openml/datasets/dataset.py

@@ -91,7 +91,10 @@ def __init__(self, dataset_id=None, name=None, version=None, description=None,
 
         if data_file is not None:
             if self._data_features_supported():
-                self.data_pickle_file = data_file.replace('.arff', '.pkl')
+                if six.PY2:
+                    self.data_pickle_file = data_file.replace('.arff', '.pkl.py2')
+                else:
+                    self.data_pickle_file = data_file.replace('.arff', '.pkl.py3')
 
                 if os.path.exists(self.data_pickle_file):
                     logger.debug("Data pickle file already exists.")

openml/tasks/split.py

@@ -1,6 +1,6 @@
 from collections import namedtuple, OrderedDict
 import os
-import sys
+import six
 
 import numpy as np
 import scipy.io.arff
@@ -60,11 +60,18 @@ def __eq__(self, other):
     @classmethod
     def _from_arff_file(cls, filename, cache=True):
         repetitions = None
-        pkl_filename = filename.replace(".arff", ".pkl")
+        if six.PY2:
+            pkl_filename = filename.replace(".arff", ".pkl.py2")
+        else:
+            pkl_filename = filename.replace(".arff", ".pkl.py3")
         if cache:
             if os.path.exists(pkl_filename):
-                with open(pkl_filename, "rb") as fh:
-                    _ = pickle.load(fh)
+                try:
+                    with open(pkl_filename, "rb") as fh:
+                        _ = pickle.load(fh)
+                except UnicodeDecodeError as e:
+                    # Possibly pickle file was created with python2 and python3 is being used to load the data
+                    raise e
                 repetitions = _["repetitions"]
                 name = _["name"]
 

tests/test_tasks/test_task_functions.py

@@ -111,6 +111,10 @@ def test_list_tasks_per_type_paginate(self):
     def test__get_task(self):
         openml.config.set_cache_directory(self.static_cache_dir)
         task = openml.tasks.get_task(1882)
+        # Test the following task as it used to throw an Unicode Error.
+        # https://github.com/openml/openml-python/issues/378
+        openml.config.server = self.production_server
+        production_task = openml.tasks.get_task(34536)
 
     def test_get_task(self):
         task = openml.tasks.get_task(1)