def make_critic(
self,
features_extractor: Optional[BaseFeaturesExtractor] = None
) -> ContinuousCritic:
critic_kwargs = self._update_features_extractor(
self.critic_kwargs, features_extractor)
return ContinuousCritic(**critic_kwargs).to(self.device)
def __init__(self,
observation_space: spaces.Space,
real_observation_space: spaces.Space,
action_space: spaces.Space,
real_action_space: spaces.Space,
net_arch: List[int],
features_extractor: nn.Module,
features_dim: int,
activation_fn: Type[nn.Module] = nn.ReLU,
normalize_images: bool = True,
n_critics: int = 2,
share_features_extractor: bool = False,
agent_num: int = 1):
""" Constructor
Args:
observation_space: Obervation space
real_observation_space: The observation space for EACH AGENT
action_space: Action space
real_action_spae: The action space for EACH AGENT
net_arch: Network architecture
features_extractor: Network to extract features
(a CNN when using images, a nn.Flatten() layer otherwise)
features_dim: Number of features
activation_fn: Activation function
normalize_images: Whether to normalize images or not,
dividing by 255.0 (True by default)
n_critics: Number of critic networks to create.
share_features_extractor: Whether the features extractor is shared or not
between the actor and the critic (this saves computation time)
"""
super().__init__(
observation_space,
action_space,
features_extractor=features_extractor,
normalize_images=normalize_images,
)
self._critics = [
ContinuousCritic(observation_space=real_observation_space,
action_space=action_space,
net_arch=net_arch,
features_extractor=features_extractor,
features_dim=features_dim,
activation_fn=activation_fn,
normalize_images=normalize_images,
n_critics=n_critics,
share_features_extractor=False)
for _ in range(agent_num)
]
for idx, criticNN in enumerate(self._critics):
self.add_module(f"ag_c_{idx}", criticNN)
def make_critic(self) -> ContinuousCritic:
return ContinuousCritic(**self.critic_kwargs).to(self.device)