def after_init(policy: Policy, obs_space: gym.Space, action_space: gym.Space,
config: TrainerConfigDict) -> None:
ComputeTDErrorMixin.__init__(policy)
TargetNetworkMixin.__init__(policy, obs_space, action_space, config)
# Move target net to device (this is done autoatically for the
# policy.model, but not for any other models the policy has).
policy.target_q_model = policy.target_q_model.to(policy.device)
def before_loss_init(
policy: Policy,
obs_space: gym.spaces.Space,
action_space: gym.spaces.Space,
config: TrainerConfigDict,
) -> None:
ComputeTDErrorMixin.__init__(policy)
TargetNetworkMixin.__init__(policy)
def setup_late_mixins(policy: Policy, obs_space: gym.spaces.Space,
action_space: gym.spaces.Space,
config: TrainerConfigDict):
"""Call all mixin classes' constructors after APPOPolicy initialization.
Args:
policy (Policy): The Policy object.
obs_space (gym.spaces.Space): The Policy's observation space.
action_space (gym.spaces.Space): The Policy's action space.
config (TrainerConfigDict): The Policy's config.
"""
KLCoeffMixin.__init__(policy, config)
ValueNetworkMixin.__init__(policy, obs_space, action_space, config)
TargetNetworkMixin.__init__(policy)
