ActionRecordWorker

class maze.core.rollout.action_record_rollout_runner.ActionRecordWorker

Class encapsulating functionality performed in worker processes.

static run(env_config: omegaconf.DictConfig, wrapper_config: omegaconf.DictConfig, agent_config: omegaconf.DictConfig, deterministic: bool, max_episode_steps: int, record_trajectory: bool, input_directory: str, reporting_queue: multiprocessing.context.BaseContext.Queue, seeding_queue: multiprocessing.context.BaseContext.Queue) → None

Build the environment and run the rollout for the specified number of episodes.

Parameters

• env_config – Hydra configuration of the environment to instantiate.

• wrapper_config – Hydra configuration of environment wrappers.

• agent_config – Hydra configuration of agent’s policies.

• deterministic – Deterministic or stochastic action sampling.

• max_episode_steps – Max number of steps per episode to perform (episode might end earlier if env returns done).

• record_trajectory – Whether to record trajectory data.

• input_directory – Directory to load the model from.

• reporting_queue – Queue for passing the stats and event logs back to the main process after each episode.

• seeding_queue – Queue for retrieving seeds.