def add_policy(
self, parsed_behavior_id: BehaviorIdentifiers, policy: Policy
) -> None:
"""
Adds policy to trainer.
:param parsed_behavior_id: Behavior identifiers that the policy should belong to.
:param policy: Policy to associate with name_behavior_id.
"""
if self.policy:
logger.warning(
"Your environment contains multiple teams, but {} doesn't support adversarial games. Enable self-play to \
train adversarial games.".format(
self.__class__.__name__
)
)
self.policy = policy
self.policies[parsed_behavior_id.behavior_id] = policy
self.optimizer = self.create_ppo_optimizer()
for _reward_signal in self.optimizer.reward_signals.keys():
self.collected_rewards[_reward_signal] = defaultdict(lambda: 0)
self.model_saver.register(self.policy)
self.model_saver.register(self.optimizer)
self.model_saver.initialize_or_load()
# Needed to resume loads properly
self.step = policy.get_current_step()
def add_policy(self, parsed_behavior_id: BehaviorIdentifiers,
policy: Policy) -> None:
"""
Adds policy to trainer.
"""
if self.policy:
logger.warning(
"Your environment contains multiple teams, but {} doesn't support adversarial games. Enable self-play to \
train adversarial games.".format(self.__class__.__name__))
self.policy = policy
self.policies[parsed_behavior_id.behavior_id] = policy
self.optimizer = self.create_sac_optimizer()
for _reward_signal in self.optimizer.reward_signals.keys():
self.collected_rewards[_reward_signal] = defaultdict(lambda: 0)
self.model_saver.register(self.policy)
self.model_saver.register(self.optimizer)
self.model_saver.initialize_or_load()
# Needed to resume loads properly
self._step = policy.get_current_step()
# Assume steps were updated at the correct ratio before
self.update_steps = int(max(1, self._step / self.steps_per_update))
self.reward_signal_update_steps = int(
max(1, self._step / self.reward_signal_steps_per_update))
def add_policy(self, parsed_behavior_id: BehaviorIdentifiers,
policy: Policy) -> None:
"""
Adds policy to trainer.
:param parsed_behavior_id: Behavior identifiers that the policy should belong to.
:param policy: Policy to associate with name_behavior_id.
"""
if not isinstance(policy, TorchPolicy):
raise RuntimeError(
f"policy {policy} must be an instance of TorchPolicy.")
self.policy = policy
self.policies[parsed_behavior_id.behavior_id] = policy
self.optimizer = self.create_poca_optimizer()
for _reward_signal in self.optimizer.reward_signals.keys():
self.collected_rewards[_reward_signal] = defaultdict(lambda: 0)
self.model_saver.register(self.policy)
self.model_saver.register(self.optimizer)
self.model_saver.initialize_or_load()
# Needed to resume loads properly
self.step = policy.get_current_step()