def _serialize_hyperparameter(self, hyperparameter_name):
"""Serialize a hyperparameter that can be a float, callable, or Tensor."""
value = self._hyper[hyperparameter_name]
if isinstance(value, learning_rate_schedule.LearningRateSchedule):
return learning_rate_schedule.serialize(value)
if callable(value):
return value()
if tensor_util.is_tensor(value):
return backend.get_value(value)
return value
def _serialize_hyperparameter(self, hyperparameter_name):
"""Serialize a hyperparameter that can be a float, callable, or Tensor."""
value = self._hyper[hyperparameter_name]
if isinstance(value, learning_rate_schedule.LearningRateSchedule):
return learning_rate_schedule.serialize(value)
if callable(value):
return value()
if tensor_util.is_tensor(value):
return backend.get_value(value)
return value
def _serialize_hyperparameter(self, hyperparameter_name):
"""Serialize a hyperparameter that can be a float, callable, or Tensor."""
value = self._hyper[hyperparameter_name]
if isinstance(value, learning_rate_schedule.LearningRateSchedule):
return learning_rate_schedule.serialize(value)
if callable(value):
return value()
if isinstance(value, (ops.Tensor, tf_variables.Variable,
distributed_values.TPUMirroredVariable)):
return backend.get_value(value)
return value
def _serialize_hyperparameter(self, hyperparameter_name):
"""Serialize a hyperparameter that can be a float, callable, or Tensor."""
value = self._hyper[hyperparameter_name]
if isinstance(value, learning_rate_schedule.LearningRateSchedule):
return learning_rate_schedule.serialize(value)
if callable(value):
return value()
if tensor_util.is_tensor(value):
return backend.get_value(value)
return value
Args:
var_list: A list of `Variable` objects.
state: An object with these methods:
`create_slot(var, val, slot_name, optional_op_name)`,
`create_slot_with_initializer(`
`var, initializer, shape, dtype, slot_name, optional_op_name)`,
`zeros_slot(var, slot_name, optional_op_name)`,
`create_non_slot_variable(initial_value, name, colocate_with)`,
`get_hyper(name)`
def _maybe_serialized(lr_decay, serialize_and_deserialize):
if serialize_and_deserialize:
serialized = learning_rate_schedule.serialize(lr_decay)
return learning_rate_schedule.deserialize(serialized)
else:
return lr_decay
def _maybe_serialized(lr_decay, serialize_and_deserialize):
if serialize_and_deserialize:
serialized = learning_rate_schedule.serialize(lr_decay)
return learning_rate_schedule.deserialize(serialized)
else:
return lr_decay
