def __init__(self, cost, parameters, update_equation, extra_layers=None):
if not isinstance(parameters, v2_parameters.Parameters):
raise TypeError('parameters should be parameters')
if not isinstance(update_equation, v2_optimizer.Optimizer):
raise TypeError("update equation parameter must be "
"paddle.v2.optimizer.Optimizer")
topology = Topology(cost, extra_layers=extra_layers)
self.__optimizer__ = update_equation
self.__topology__ = topology
self.__parameters__ = parameters
self.__topology_in_proto__ = topology.proto()
# In local mode, disable sparse_remote_update.
for param in self.__topology_in_proto__.parameters:
if param.sparse_remote_update:
param.sparse_remote_update = False
self.__data_types__ = topology.data_type()
gm = api.GradientMachine.createFromConfigProto(
self.__topology_in_proto__, api.CREATE_MODE_NORMAL,
self.__optimizer__.enable_types())
assert isinstance(gm, api.GradientMachine)
self.__gradient_machine__ = gm
self.__gradient_machine__.randParameters()
parameters.append_gradient_machine(gm)
def __init__(self,
cost,
parameters,
update_equation,
extra_layers=None,
is_local=True,
pserver_spec=None,
use_etcd=True):
if not isinstance(parameters, v2_parameters.Parameters):
raise TypeError('parameters should be parameters')
if not isinstance(update_equation, v2_optimizer.Optimizer):
raise TypeError("update equation parameter must be "
"paddle.v2.optimizer.Optimizer")
import py_paddle.swig_paddle as api
topology = Topology(cost, extra_layers=extra_layers)
# HACK(typhoonzero): update ParameterConfig(proto) in case of optimizers
# are defined after layers, or between layers.
topology.update_from_default()
parameters.update_param_conf(topology.proto())
self.__optimizer__ = update_equation
self.__topology__ = topology
self.__parameters__ = parameters
self.__topology_in_proto__ = topology.proto()
self.__is_local__ = is_local
self.__pserver_spec__ = pserver_spec
self.__use_etcd__ = use_etcd
self.__use_sparse_updater__ = self.__topology__.use_sparse_updater()
# # In local mode, disable sparse_remote_update.
if is_local:
for param in self.__topology_in_proto__.parameters:
if param.sparse_remote_update:
param.sparse_remote_update = False
self.__gm_create_mode__ = api.CREATE_MODE_NORMAL if not \
self.__use_sparse_updater__ else api.CREATE_MODE_SGD_SPARSE_CPU_TRAINING
self.__data_types__ = topology.data_type()
gm = api.GradientMachine.createFromConfigProto(
self.__topology_in_proto__, self.__gm_create_mode__,
self.__optimizer__.enable_types())
assert isinstance(gm, api.GradientMachine)
self.__gradient_machine__ = gm
self.__gradient_machine__.randParameters()
self.__parameters__.append_gradient_machine(gm)
self.__parameter_updater__ = None
def __init__(self,
cost,
parameters,
update_equation,
extra_layers=None,
is_local=True,
pserver_spec=None,
use_etcd=True):
if not isinstance(parameters, v2_parameters.Parameters):
raise TypeError('parameters should be parameters')
if not isinstance(update_equation, v2_optimizer.Optimizer):
raise TypeError("update equation parameter must be "
"paddle.v2.optimizer.Optimizer")
import py_paddle.swig_paddle as api
topology = Topology(cost, extra_layers=extra_layers)
# HACK(typhoonzero): update ParameterConfig(proto) in case of optimizers
# are defined after layers, or between layers.
topology.update_from_default()
parameters.update_param_conf(topology.proto())
self.__optimizer__ = update_equation
self.__topology__ = topology
self.__parameters__ = parameters
self.__topology_in_proto__ = topology.proto()
self.__is_local__ = is_local
self.__pserver_spec__ = pserver_spec
self.__use_etcd__ = use_etcd
self.__use_sparse_updater__ = self.__topology__.use_sparse_updater()
# # In local mode, disable sparse_remote_update.
if is_local:
for param in self.__topology_in_proto__.parameters:
if param.sparse_remote_update:
param.sparse_remote_update = False
self.__gm_create_mode__ = api.CREATE_MODE_NORMAL if not \
self.__use_sparse_updater__ else api.CREATE_MODE_SGD_SPARSE_CPU_TRAINING
self.__data_types__ = topology.data_type()
gm = api.GradientMachine.createFromConfigProto(
self.__topology_in_proto__, self.__gm_create_mode__,
self.__optimizer__.enable_types())
assert isinstance(gm, api.GradientMachine)
self.__gradient_machine__ = gm
self.__gradient_machine__.randParameters()
self.__parameters__.append_gradient_machine(gm)
self.__parameter_updater__ = None
