def test_deserialization(self):
serialized_feature_columns = feature.serialize_feature_columns(
self._feature_columns)
restored_feature_columns = feature.deserialize_feature_columns(
serialized_feature_columns, custom_objects=self._custom_objects)
self.assertEqual(restored_feature_columns['utility'].get_config(),
self._feature_columns['utility'].get_config())
self.assertEqual(restored_feature_columns['unigrams'].get_config(),
self._feature_columns['unigrams'].get_config())
def from_config(cls, config, custom_objects=None):
"""Creates a RankingNetwork layer from its config.
Args:
config: (dict) Layer configuration, typically the output of `get_config`.
custom_objects: (dict) Optional dictionary mapping names to custom classes
or functions to be considered during deserialization.
Returns:
A RankingNetwork layer.
"""
config_cp = config.copy()
config_cp[
'context_feature_columns'] = feature.deserialize_feature_columns(
config_cp['context_feature_columns'],
custom_objects=custom_objects)
config_cp[
'example_feature_columns'] = feature.deserialize_feature_columns(
config_cp['example_feature_columns'],
custom_objects=custom_objects)
return cls(**config_cp)
def test_deserialization(self):
serialized_feature_columns = feature.serialize_feature_columns(
self._feature_columns)
restored_feature_columns = feature.deserialize_feature_columns(
serialized_feature_columns, custom_objects=self._custom_objects)
self.assertEqual(restored_feature_columns['utility'],
self._feature_columns['utility'])
# TODO: Deserialized embedding feature column behavior is the
# same but config is different. Hence we check for individual attributes.
self.assertEqual(restored_feature_columns['unigrams'].name,
'unigrams_embedding')
self.assertEqual(restored_feature_columns['unigrams'].initializer.mean,
0.0)
self.assertCountEqual(
restored_feature_columns['unigrams'].categorical_column.
vocabulary_list,
['ranking', 'regression', 'classification', 'ordinal'])