def add_policy(self, name_behavior_id: str, policy: TFPolicy) -> None:
"""
Adds policy to trainer.
:param brain_parameters: specifications for policy construction
"""
if self.policy:
logger.warning(
"add_policy has been called twice. {} is not a multi-agent trainer"
.format(self.__class__.__name__))
if not isinstance(policy, SACPolicy):
raise RuntimeError(
"Non-SACPolicy passed to SACTrainer.add_policy()")
self.policy = policy
self.step = policy.get_current_step()
def add_policy(self, name_behavior_id: str, policy: TFPolicy) -> None:
"""
Adds policy to trainer.
:param brain_parameters: specifications for policy construction
"""
if self.policy:
logger.warning(
"add_policy has been called twice. {} is not a multi-agent trainer"
.format(self.__class__.__name__))
if not isinstance(policy, NNPolicy):
raise RuntimeError(
"Non-SACPolicy passed to SACTrainer.add_policy()")
self.policy = policy
self.optimizer = SACOptimizer(self.policy, self.trainer_parameters)
for _reward_signal in self.optimizer.reward_signals.keys():
self.collected_rewards[_reward_signal] = defaultdict(lambda: 0)
# Needed to resume loads properly
self.step = policy.get_current_step()
self.next_summary_step = self._get_next_summary_step()
def add_policy(self, name_behavior_id: str, policy: TFPolicy) -> None:
"""
Adds policy to trainer.
:param name_behavior_id: Behavior ID that the policy should belong to.
:param policy: Policy to associate with name_behavior_id.
"""
if self.policy:
logger.warning(
"add_policy has been called twice. {} is not a multi-agent trainer"
.format(self.__class__.__name__))
if not isinstance(policy, NNPolicy):
raise RuntimeError(
"Non-NNPolicy passed to PPOTrainer.add_policy()")
self.policy = policy
self.optimizer = PPOOptimizer(self.policy, self.trainer_parameters)
for _reward_signal in self.optimizer.reward_signals.keys():
self.collected_rewards[_reward_signal] = defaultdict(lambda: 0)
# Needed to resume loads properly
self.step = policy.get_current_step()
self.next_summary_step = self._get_next_summary_step()