| import torch | |
| class RandomActor: | |
| """Random actor. | |
| Args: | |
| env: Environment. | |
| is_controlled_func (torch.Tensor): Determines which agents are controlled by this actor. Shape: (max_num_agents,). | |
| valid_agent_mask (torch.Tensor): Mask that determines which agents are valid, and thus controllable, in the environment. Shape: (num_worlds, max_num_agents). | |
| device (str): Device to put the actions on. | |
| """ | |
| def __init__( | |
| self, env, is_controlled_func, valid_agent_mask, device="cuda" | |
| ): | |
| self.env = env | |
| self.is_controlled_func = is_controlled_func | |
| self.device = device | |
| self.valid_and_controlled_mask = self.get_valid_actor_mask( | |
| is_controlled_func, valid_agent_mask | |
| ) | |
| self.actor_ids = [ | |
| torch.where(self.valid_and_controlled_mask[world_idx, :])[0] | |
| for world_idx in range(valid_agent_mask.shape[0]) | |
| ] | |
| def select_action(self): | |
| """Select random actions.""" | |
| action_lists = [] | |
| for world_idx in range(len(self.actor_ids)): | |
| actions = torch.Tensor( | |
| [ | |
| self.env.action_space.sample() | |
| for _ in range(len(self.actor_ids[world_idx])) | |
| ] | |
| ).to(self.device) | |
| action_lists.append(actions) | |
| return action_lists | |
| def get_valid_actor_mask(self, is_controlled_func, valid_agent_mask): | |
| """Returns a boolean mask across worlds that indicates which agents | |
| are valid _and_ controlled by this actor. | |
| """ | |
| num_worlds = valid_agent_mask.shape[0] | |
| is_controlled_func = is_controlled_func.expand((num_worlds, -1)) | |
| return is_controlled_func.to(self.device) & valid_agent_mask.to( | |
| self.device | |
| ) | |