def sync_target_network(self) -> None:
"""Synchronize target network with current network."""
synchronize_parameters(
src=self.model.q_function,
dst=self.target_model,
method=self.target_update_method,
tau=self.soft_update_tau,
)
def sync_target_network(self):
"""Synchronize target network with current network."""
synchronize_parameters(
src=self.q_func1,
dst=self.target_q_func1,
method="soft",
tau=self.soft_update_tau,
)
synchronize_parameters(
src=self.q_func2,
dst=self.target_q_func2,
method="soft",
tau=self.soft_update_tau,
)
def sync_target_network(self):
"""Synchronize target network with current network."""
if self.target_model is None:
self.target_model = copy.deepcopy(self.model)
def flatten_parameters(mod):
if isinstance(mod, torch.nn.RNNBase):
mod.flatten_parameters()
# RNNBase.flatten_parameters must be called again after deep-copy.
# See: https://discuss.pytorch.org/t/why-do-we-need-flatten-parameters-when-using-rnn-with-dataparallel/46506 # NOQA
self.target_model.apply(flatten_parameters)
# set target n/w to evaluate only.
self.target_model.eval()
else:
synchronize_parameters(
src=self.model,
dst=self.target_model,
method=self.target_update_method,
tau=self.soft_update_tau,
)
