def get_config(self):
module = self.function.__module__
if isinstance(self.function, python_types.LambdaType):
function = generic_utils.func_dump(self.function)
function_type = 'lambda'
else:
function = self.function.__name__
function_type = 'function'
output_shape_module = None
if isinstance(self._output_shape, python_types.LambdaType):
output_shape = generic_utils.func_dump(self._output_shape)
output_shape_type = 'lambda'
output_shape_module = self._output_shape.__module__
elif callable(self._output_shape):
output_shape = self._output_shape.__name__
output_shape_type = 'function'
output_shape_module = self._output_shape.__module__
else:
output_shape = self._output_shape
output_shape_type = 'raw'
config = {
'function': function,
'module': module,
'function_type': function_type,
'output_shape': output_shape,
'output_shape_type': output_shape_type,
'output_shape_module': output_shape_module,
'arguments': self.arguments
}
base_config = super(Lambda, self).get_config()
return dict(list(base_config.items()) + list(config.items()))
def get_config(self):
module = self.function.__module__
if isinstance(self.function, python_types.LambdaType):
function = generic_utils.func_dump(self.function)
function_type = 'lambda'
else:
function = self.function.__name__
function_type = 'function'
output_shape_module = None
if isinstance(self._output_shape, python_types.LambdaType):
output_shape = generic_utils.func_dump(self._output_shape)
output_shape_type = 'lambda'
output_shape_module = self._output_shape.__module__
elif callable(self._output_shape):
output_shape = self._output_shape.__name__
output_shape_type = 'function'
output_shape_module = self._output_shape.__module__
else:
output_shape = self._output_shape
output_shape_type = 'raw'
config = {
'function': function,
'module': module,
'function_type': function_type,
'output_shape': output_shape,
'output_shape_type': output_shape_type,
'output_shape_module': output_shape_module,
'arguments': self.arguments
}
base_config = super(Lambda, self).get_config()
return dict(list(base_config.items()) + list(config.items()))
def _serialize_function_to_config(function):
"""Serialize the function for get_config()."""
if isinstance(function, python_types.LambdaType):
output = generic_utils.func_dump(function)
output_type = "lambda"
module = function.__module__
elif callable(function):
output = function.__name__
output_type = "function"
module = function.__module__
else:
raise ValueError("Unrecognized function type for input: {}".format(
type(function)))
return output, output_type, module
def get_config(self):
configs = super(Network, self).get_config()
fn = self.fn_layer_creator
if isinstance(fn, python_types.LambdaType):
fn = (generic_utils.func_dump(fn), 'lambda', fn.__module__)
elif callable(fn):
fn = (fn.__name__, 'function', fn.__module__)
configs.update({
'fn_layer_creator': fn,
'delay_context': self.delay_context,
'pooling': self.pooling,
})
return configs
def _serialize_function_to_config(self, inputs, allow_raw=False):
if isinstance(inputs, python_types.LambdaType):
output = generic_utils.func_dump(inputs)
output_type = 'lambda'
module = inputs.__module__
elif callable(inputs):
output = inputs.__name__
output_type = 'function'
module = inputs.__module__
elif allow_raw:
output = inputs
output_type = 'raw'
module = None
else:
raise ValueError('Invalid input for serialization, type: %s ' %
type(inputs))
return output, output_type, module
def serialize_function(func):
"""Serializes function for Keras.
(De)serializing Python functions from/to bytecode is unsafe. Therefore we
return the function's type as an anonymous function ('lambda') or named
function in the Python environment ('function'). In the latter case, this lets
us use the Python scope to obtain the function rather than reload it from
bytecode. (Note that both cases are brittle!)
This serialization mimicks the implementation in `tf.keras.layers.Lambda`.
Args:
func: Python function to serialize.
Returns:
(serial, function_type): Serialized object, which is a tuple of its
bytecode (if function is anonymous) or name (if function is named), and its
function type.
"""
if isinstance(func, types.LambdaType):
return generic_utils.func_dump(func), 'lambda'
return func.__name__, 'function'
def serialize_function(func):
"""Serializes function for Keras.
(De)serializing Python functions from/to bytecode is unsafe. Therefore we
return the function's type as an anonymous function ('lambda') or named
function in the Python environment ('function'). In the latter case, this lets
us use the Python scope to obtain the function rather than reload it from
bytecode. (Note that both cases are brittle!)
This serialization mimicks the implementation in `tf.keras.layers.Lambda`.
Args:
func: Python function to serialize.
Returns:
(serial, function_type): Serialized object, which is a tuple of its
bytecode (if function is anonymous) or name (if function is named), and its
function type.
"""
if isinstance(func, types.LambdaType):
return generic_utils.func_dump(func), 'lambda'
return func.__name__, 'function'
