def from_base_network(cls,
base_network,
json_content=None,
share_params=False,
base_as_calc_step=False,
**kwargs):
"""
:param LayerNetwork base_network: base network to derive from
:param dict[str]|None json_content: JSON content for subnetwork. if None, will use from base network
:param bool share_params: will use the same params as the base network
:param bool base_as_calc_step: base is calc step 0. see below
:param dict[str] kwargs: kwargs for __init__
:rtype: LayerNetwork
"""
if "n_out" in kwargs and "n_in" not in kwargs:
kwargs["n_in"] = None
network = cls(
base_network=base_network,
shared_params_network=base_network if share_params else None,
**dict_joined(base_network.init_args(), kwargs))
if base_as_calc_step:
network.calc_step_base = base_network # used by CalcStepLayer. see also get_calc_step()
if json_content is None:
json_content = base_network.to_json_content()
cls.from_json(json_content, network=network)
if share_params:
trainable_params = network.get_all_params_vars()
assert len(trainable_params) == 0
return network
def from_json_and_config(cls, json_content, config, **kwargs):
"""
:type config: Config.Config
:type json_content: str | dict
:rtype: LayerNetwork
"""
network = cls.from_json(json_content, **dict_joined(kwargs, cls.init_args_from_config(config)))
network.recurrent = network.recurrent or config.bool('recurrent', False)
return network
def from_json_and_config(cls, json_content, config, **kwargs):
"""
:type config: Config.Config
:type json_content: str | dict
:rtype: LayerNetwork
"""
network = cls.from_json(json_content, **dict_joined(kwargs, cls.init_args_from_config(config)))
network.recurrent = network.recurrent or config.bool('recurrent', False)
return network
def from_base_network(cls, base_network, json_content=None, share_params=False, base_as_calc_step=False, **kwargs):
"""
:param LayerNetwork base_network: base network to derive from
:param dict[str]|None json_content: JSON content for subnetwork. if None, will use from base network
:param bool share_params: will use the same params as the base network
:param bool base_as_calc_step: base is calc step 0. see below
:param dict[str] kwargs: kwargs for __init__
:rtype: LayerNetwork
"""
network = cls(
base_network=base_network,
shared_params_network=base_network if share_params else None,
**dict_joined(base_network.init_args(), kwargs))
if base_as_calc_step:
network.calc_step_base = base_network # used by CalcStepLayer. see also get_calc_step()
if json_content is None:
json_content = base_network.to_json_content()
cls.from_json(json_content, network=network)
if share_params:
trainable_params = network.get_all_params_vars()
assert len(trainable_params) == 0
return network