def _get_sequence_dense_tensor_state(column, features):
state_manager = fc._StateManagerImpl(fc_lib.DenseFeatures(column),
trainable=True)
column.create_state(state_manager)
dense_tensor, lengths = column.get_sequence_dense_tensor(
fc.FeatureTransformationCache(features), state_manager)
return dense_tensor, lengths, state_manager
def fc2_implements_resources():
"""Returns true if imported TF version implements resources for FCv2."""
if not hasattr(feature_column_v2, "DenseColumn"):
return False
if not hasattr(feature_column_v2, "_StateManagerImpl"):
return False
state_manager = feature_column_v2._StateManagerImpl( # pylint: disable=protected-access
layer=None, trainable=False)
try:
state_manager.add_resource("COLUMN_DUMMY", "RESOURCE_DUMMY", True)
except NotImplementedError:
return False
return True
def __init__(self,
feature_columns,
expected_column_type,
trainable,
name,
partitioner=None,
**kwargs):
super(_BaseFeaturesLayer, self).__init__(
name=name, trainable=trainable, **kwargs)
self._feature_columns = feature_column_v2._normalize_feature_columns( # pylint: disable=protected-access
feature_columns)
self._state_manager = feature_column_v2._StateManagerImpl( # pylint: disable=protected-access
self, self.trainable)
self._partitioner = partitioner
for column in self._feature_columns:
if not isinstance(column, expected_column_type):
raise ValueError(
'Items of feature_columns must be a {}. '
'You can wrap a categorical column with an '
'embedding_column or indicator_column. Given: {}'.format(
expected_column_type, column))