ADD serialize cv objects as parameters

Author: mfeurer

openml/flows/sklearn_converter.py

@@ -114,7 +114,8 @@ def flow_to_sklearn(o, **kwargs):
                     rval = component
                 else:
                     rval = (step_name, component)
-
+            elif serialized_type == 'cv_object':
+                rval = _deserialize_cross_validator(value, **kwargs)
             else:
                 raise ValueError('Cannot flow_to_sklearn %s' % serialized_type)
 
@@ -401,12 +402,10 @@ def deserialize_function(name, **kwargs):
         return None
     return function_handle
 
-# This produces a flow, thus it does not need a deserialize function as
-# the function _deserialize_model is used for that. It cannot be fed
-# to serialize_model() because cross-validators do not have get_params().
-def _serialize_cross_validator( o):
+def _serialize_cross_validator(o):
+    ret = OrderedDict()
+
     parameters = OrderedDict()
-    parameters_meta_info = OrderedDict()
 
     # XXX this is copied from sklearn.model_selection._split
     cls = o.__class__
@@ -440,26 +439,25 @@ def _serialize_cross_validator( o):
             parameters[key] = value
         else:
             parameters[key] = None
-        parameters_meta_info[key] = OrderedDict((('description', None),
-                                                 ('data_type', None)))
 
-    # Create a flow
+    ret['oml:serialized_object'] = 'cv_object'
     name = o.__module__ + "." + o.__class__.__name__
+    value = OrderedDict(name=name, parameters=parameters)
+    ret['value'] = value
 
-    external_version = _get_external_version_info()
-    flow = OpenMLFlow(name=name,
-                      description='Automatically created sub-component.',
-                      model=o,
-                      parameters=parameters,
-                      parameters_meta_info=parameters_meta_info,
-                      external_version=external_version,
-                      components=OrderedDict(),
-                      tags=[],
-                      language='English',
-                      # TODO fill in dependencies!
-                      dependencies=None)
+    return ret
 
-    return flow
+
+def _deserialize_cross_validator(value, **kwargs):
+    model_name = value['name']
+    parameters = value['parameters']
+
+    module_name = model_name.rsplit('.', 1)
+    model_class = getattr(importlib.import_module(module_name[0]),
+                          module_name[1])
+    for parameter in parameters:
+        parameters[parameter] = flow_to_sklearn(parameters[parameter])
+    return model_class(**parameters)
 
 
 def _get_external_version_info():

tests/flows/test_flow.py

@@ -198,7 +198,6 @@ def test_sklearn_to_upload_to_flow(self):
         self.assertIsNot(new_flow, flow)
 
         fixture_name = 'sklearn.model_selection._search.RandomizedSearchCV(' \
-                       'cv=sklearn.model_selection._split.StratifiedKFold,' \
                        'estimator=sklearn.pipeline.Pipeline(' \
                        'sklearn.preprocessing.data.OneHotEncoder,' \
                        'sklearn.preprocessing.data.StandardScaler,' \

tests/flows/test_sklearn.py

@@ -225,7 +225,6 @@ def test_serialize_complex_flow(self):
         serialized = sklearn_to_flow(rs)
 
         fixture_name = 'sklearn.model_selection._search.RandomizedSearchCV(' \
-                       'cv=sklearn.model_selection._split.StratifiedKFold,' \
                        'estimator=sklearn.pipeline.Pipeline(' \
                        'sklearn.preprocessing.data.OneHotEncoder,' \
                        'sklearn.preprocessing.data.StandardScaler,' \
@@ -269,6 +268,25 @@ def test_serialize_function(self):
         deserialized = flow_to_sklearn(serialized)
         self.assertEqual(deserialized, sklearn.feature_selection.chi2)
 
+    def test_serialize_cvobject(self):
+        methods = [sklearn.model_selection.KFold(3),
+                   sklearn.model_selection.LeaveOneOut()]
+        fixtures = [OrderedDict([('oml:serialized_object', 'cv_object'),
+                                 ('value', OrderedDict([('name', 'sklearn.model_selection._split.KFold'),
+                                                        ('parameters', OrderedDict([('n_splits', '3'),
+                                                                                    ('random_state', 'null'),
+                                                                                    ('shuffle', 'false')]))]))]),
+                    OrderedDict([('oml:serialized_object', 'cv_object'),
+                                 ('value', OrderedDict([('name', 'sklearn.model_selection._split.LeaveOneOut'),
+                                                        ('parameters', OrderedDict())]))])]
+        for method, fixture in zip(methods, fixtures):
+            m = sklearn_to_flow(method)
+            self.assertEqual(m, fixture)
+
+            m_new = flow_to_sklearn(m)
+            self.assertIsNot(m_new, m)
+            self.assertIsInstance(m_new, type(method))
+
     def test_serialize_simple_parameter_grid(self):
         # TODO instead a GridSearchCV object should be serialized