Use new OpenML API

Author: mfeurer

openml/apiconnector.py

@@ -27,7 +27,7 @@
 
 logger = logging.getLogger(__name__)
 
-OPENML_URL = "http://www.openml.org"
+OPENML_URL = "http://api_new.openml.org/v1/"
 
 
 class OpenMLStatusChange(Warning):
@@ -50,10 +50,6 @@ def __init__(self, message):
         super(OpenMLServerError, self).__init__(message)
 
 
-class AuthentificationError(PyOpenMLError):
-    def __init__(self, message):
-        super(AuthentificationError, self).__init__(message)
-
 class OpenMLCacheException(PyOpenMLError):
     def __init__(self, message):
         super(OpenMLCacheException, self).__init__(message)
@@ -77,14 +73,9 @@ class APIConnector(object):
         directory '.openml/cache' in the users home directory will be used.
         If either directory does not exist, it will be created.
 
-    username : string, optional (default=None)
-        Your username.
-
-    password : string, optional (default=None)
-        Your passwort. This will not be stored! Instead, the md5 hash is
-        calculated and used to authenticate to the OpenML server,
-        which returns a session key. This session key is the only credential
-        which is stored.
+    apikey : string, optional (default=None)
+        Your OpenML API key which will be used to authenticate you at the OpenML
+        server.
 
     server : string, optional (default=None)
         The OpenML server to connect to.
@@ -110,24 +101,18 @@ class APIConnector(object):
     Raises
     ------
     ValueError
-        If one of the following variables is neither specified in the
-        configuration file nor when creating the APIconnector class:
-        username, password
-    AuthentificationError
-        If authentification at the OpenML server does not work.
+        If apikey is neither specified in the config nor given as an argument.
     OpenMLServerError
         If the OpenML server returns an unexptected response.
 
     Testing the API calls in Firefox
     --------------------------------
-    With the Firefox AddOn HTTPRequestor, one can check the OpenML API calls
-    which need authentification (by providing such). First, create the md5
-    hash of your OpenML password and add it to
+    With the Firefox AddOn HTTPRequestor, one can check the OpenML API calls.
 
     """
-    def __init__(self, cache_directory=None, username=None, password=None,
+    def __init__(self, cache_directory=None, apikey=None,
                  server=None, verbosity=None, configure_logger=True,
-                 authenticate=True, private_directory=None):
+                 private_directory=None):
         # The .openml directory is necessary, just try to create it (EAFP)
         try:
             os.mkdir(os.path.expanduser('~/.openml'))
@@ -139,10 +124,8 @@ def __init__(self, cache_directory=None, username=None, password=None,
         self.config = self._parse_config()
         if cache_directory is not None:
             self.config.set('FAKE_SECTION', 'cachedir', cache_directory)
-        if username is not None:
-            self.config.set('FAKE_SECTION', 'username', username)
-        if password is not None:
-            self.config.set('FAKE_SECTION', 'password', password)
+        if apikey is not None:
+            self.config.set('FAKE_SECTION', 'apikey', apikey)
         if server is not None:
             self.config.set('FAKE_SECTION', 'server', server)
         if verbosity is not None:
@@ -162,11 +145,6 @@ def __init__(self, cache_directory=None, username=None, password=None,
                 format='[%(levelname)s] [%(asctime)s:%(name)s] %('
                        'message)s', datefmt='%H:%M:%S', level=level)
 
-        if authenticate:
-            self._session_hash = self._authenticate(
-                self.config.get('FAKE_SECTION', 'username'),
-                self.config.get('FAKE_SECTION', 'password'))
-
         # Set up the cache directories
         self.cache_dir = self.config.get('FAKE_SECTION', 'cachedir')
         self.dataset_cache_dir = os.path.join(self.cache_dir, "datasets")
@@ -187,50 +165,8 @@ def __init__(self, cache_directory=None, username=None, password=None,
             if not os.path.exists(dir_) and not os.path.isdir(dir_):
                 os.mkdir(dir_)
 
-    def _authenticate(self, username, password):
-        # Check the username
-        if username is None:
-            raise ValueError("No username specified.")
-        elif not is_string(username):
-            raise ValueError("Username must be of type string.")
-        elif not username:
-            raise ValueError("No value for argument username specified.")
-
-        # Check the password
-        if password is None:
-            raise ValueError("No password specified.")
-        elif not is_string(password):
-            raise ValueError("Password must be of type string.")
-        elif not password:
-            raise ValueError("No value for argument password specified.")
-
-        m = hashlib.md5()
-        m.update(password.encode('utf-8'))
-        md5 = m.hexdigest()
-
-        # TODO: catch possible exceptions
-        data = {'username': username, 'password': md5}
-        return_code, xml_string = self._perform_api_call(
-            "openml.authenticate", data=data, add_authentication=False)
-
-        xml_dict = xmltodict.parse(xml_string)
-        if xml_dict.get('oml:authenticate'):
-            session_hash = xml_dict['oml:authenticate']['oml:session_hash']
-            return session_hash
-        elif xml_dict.get('oml:error'):
-            error_code = xml_dict['oml:error']['oml:code']
-            if error_code == '252':
-                raise AuthentificationError(
-                    "Authentication failed. The username and password did not "
-                    "match any record in the database.")
-            else:
-                OpenMLServerError(
-                    "Unexpected server response code %d with response"
-                    " message %s" % (return_code, xml_string))
-
     def _parse_config(self):
-        defaults = {'username': '',
-                    'password': '',
+        defaults = {'apikey': '',
                     'server': OPENML_URL,
                     'verbosity': 0,
                     'cachedir': os.path.expanduser('~/.openml/cache'),
@@ -436,7 +372,7 @@ def get_dataset_list(self):
             these are also returned.
         """
         # TODO add proper error handling here!
-        return_code, xml_string = self._perform_api_call("openml.data")
+        return_code, xml_string = self._perform_api_call("data/list/")
         datasets_dict = xmltodict.parse(xml_string)
 
         # Minimalistic check if the XML is useful
@@ -551,7 +487,7 @@ def download_dataset_description(self, did):
         except (OpenMLCacheException):
             try:
                 return_code, dataset_xml = self._perform_api_call(
-                    "openml.data.description", data_id=did)
+                    "data/%d" % did)
             except (URLError, UnicodeEncodeError) as e:
                 # TODO logger.debug
                 self._remove_dataset_chache_dir(did)
@@ -566,7 +502,7 @@ def download_dataset_description(self, did):
                 "oml:data_set_description"]
         except Exception as e:
             # TODO logger.debug
-            self._remove_dataset_chache_dir()
+            self._remove_dataset_chache_dir(did)
             print("Dataset ID", did)
             raise e
 
@@ -611,7 +547,7 @@ def download_dataset_features(self, did):
         except (OSError, IOError):
             try:
                 return_code, features_xml = self._perform_api_call(
-                    "openml.data.features", data_id=did)
+                    "data/features/%d" % did)
             except (URLError, UnicodeEncodeError) as e:
                 # TODO logger.debug
                 print(e)
@@ -635,7 +571,7 @@ def download_dataset_qualities(self, did):
         qualities_file = os.path.join(did_cache_dir, "qualities.xml")
         try:
             return_code, qualities_xml = self._perform_api_call(
-                "openml.data.qualities", data_id=did)
+                "data/qualities/%d" % did)
         except (URLError, UnicodeEncodeError) as e:
             # TODO logger.debug
             print(e)
@@ -723,7 +659,7 @@ def get_task_list(self, task_type_id=1):
                              "cast to an Integer.")
 
         return_code, xml_string = self._perform_api_call(
-            "openml.tasks", task_type_id=task_type_id)
+            "task/list/%d" % task_type_id)
         tasks_dict = xmltodict.parse(xml_string)
         # Minimalistic check if the XML is useful
         assert tasks_dict['oml:tasks']['@xmlns:oml'] == \
@@ -773,7 +709,7 @@ def download_task(self, task_id):
 
             try:
                 return_code, task_xml = self._perform_api_call(
-                    "openml.task.search", task_id=task_id)
+                    "task/%d" % task_id)
             except (URLError, UnicodeEncodeError) as e:
                 print(e)
                 raise e
@@ -875,22 +811,18 @@ def _create_task_cache_dir(self, task_id):
             pass
         return task_cache_dir
 
-    def _perform_api_call(self, call, data=None, add_authentication=True,
-                          **kwargs):
+    def _perform_api_call(self, call, data=None):
         """Perform an API call at the OpenML server.
 
         This method must be used by all other methods using the REST API.
 
         Parameters
         ----------
         call : str
-            The API call. For example openml.task.search
+            The API call. For example data/list
         data : dict (default=None)
             Dictionary containing data which will be sent to the OpenML
             server via a POST request.
-        add_authentication : bool (default=True)
-            DO NOT CHANGE THIS. The only method which should set this
-            argument to False is `authenticate`.
         **kwargs
             Further arguments which are appended as GET arguments.
 
@@ -901,21 +833,16 @@ def _perform_api_call(self, call, data=None, add_authentication=True,
         return_value : str
             Return value of the OpenML server
         """
-        # TODO: do input validation!
-        url = self.config.get("FAKE_SECTION", "server") + "/api/?f="
-        url += "" + call
-        if kwargs:
-            for key in kwargs:
-                url += "&" + key + "=" + str(kwargs[key])
-        # TODO logger.debug(url)
-        return self._read_url(url, data=data,
-                              add_authentication=add_authentication)
-
-    def _read_url(self, url, add_authentication=False, data=None):
+        url = self.config.get("FAKE_SECTION", "server")
+        if not url.endswith("/"):
+            url += "/"
+        url += call
+        return self._read_url(url, data=data)
+
+    def _read_url(self, url, data=None):
         if data is None:
             data = {}
-        if add_authentication:
-            data = {'session_hash': self._session_hash}
+        data['session_hash'] = self.config.get('FAKE_SECTION', 'apikey')
         data = urlencode(data)
         data = data.encode('utf-8')
 

source/progress.rst

@@ -12,34 +12,50 @@ API calls
 =============================================== =========== ====== =============== ========== =====================
 API call                                        implemented tested properly tested loads json proper error handling
 =============================================== =========== ====== =============== ========== =====================
-authenticate                                    yes         yes
-authenticate.check
-data                                            yes         yes
-data.description                                yes         yes
-data.upload
-data.delete
-data.licences
-data.features                                   yes         yes
-data.qualities                                  yes         yes
-data.qualities.list
-task                                            yes         yes
-task.types.search                               yes         yes
-task.evaluations
-task.types
-estimationprocedure
-implementation.exists
-implementation.upload
-implementation.owned
-implementation.delete
-implementation.licences
-evaluation.measures
-run
-run.upload
-run.delete
-job
-setup
+/data/list/                                     yes         yes
+/data/list/active/
+/data/list/tag/{tag}
+/data/{data_id}                                 yes         yes
+/data/delete/
+/data/upload/
+/data/features/{data_id}                        yes         yes
+/data/features/upload/
+/data/qualities/{data_id}                       yes         yes
+/data/qualities/list
+/data/qualities/upload
+/data/tag
+/data/untag
+/task/list                                      yes         yes
+/task/list/active
+/task/list/tag/{tag}
+/task/{task_id}                                 yes         yes
+/task/tag
+/task/untag
+/task/delete
+/tasktype/list
+/tasktype/{task_id}
+/flow/tag
+/flow/untag
+/flow/{flow_id}
+/flow/
+/flow/exists/{name,ext_version}
+/flow/owned
+/run/list
+/run/{run_id}
+/run
+/run/tag
+/run/untag
+/run/evaluate
+/run/reset
+/estimationprocedure/{proc_id}
+/estimationprocedure/list
+/evaluationmeasures/list
+/job/request/
 =============================================== =========== ====== =============== ========== =====================
 
+This list does not contain the `/setup/` calls because we do not need them
+according to Jan.
+
 Convenience Functions
 =====================
 

tests/test_apiconnector.py

@@ -45,38 +45,15 @@ def setUp(self):
         os.chdir(self.workdir)
 
         self.cached = True
-        self.connector = APIConnector(cache_directory=self.workdir)
-        print(self.connector._session_hash)
+        self.connector = APIConnector(cache_directory=self.workdir,
+                                      apikey='test')
 
     def tearDown(self):
         os.chdir(self.cwd)
         shutil.rmtree(self.workdir)
 
     ############################################################################
     # Test administrative stuff
-    @mock.patch.object(APIConnector, '_perform_api_call', autospec=True)
-    def test_authentication(self, mock_perform_API_call):
-        # TODO return error messages
-        mock_perform_API_call.return_value = 400, \
-        """<oml:authenticate xmlns:oml = "http://openml.org/openml">
-  <oml:session_hash>G9MPPN114ZCZNWW2VN3JE9VF1FMV8Y5FXHUDUL4P</oml:session_hash>
-  <oml:valid_until>2014-08-13 20:01:29</oml:valid_until>
-  <oml:timezone>Europe/Berlin</oml:timezone>
-</oml:authenticate>"""
-
-        # This already does an authentication
-        connector = APIConnector()
-        # but we only test it here...
-        self.assertEqual(1, mock_perform_API_call.call_count)
-        self.assertEqual(connector._session_hash,
-                         "G9MPPN114ZCZNWW2VN3JE9VF1FMV8Y5FXHUDUL4P")
-
-        # Test that it actually returns what we want
-        session_hash = connector._authenticate("Bla", "Blub")
-        self.assertEqual(2, mock_perform_API_call.call_count)
-        self.assertEqual(session_hash,
-                         "G9MPPN114ZCZNWW2VN3JE9VF1FMV8Y5FXHUDUL4P")
-
     @unittest.skip("Not implemented yet.")
     def test_parse_config(self):
         raise Exception()