MultiStepActorCritic¶
-
class
maze.train.trainers.common.actor_critic.actor_critic_trainer.MultiStepActorCritic(algorithm_config: Union[maze.train.trainers.a2c.a2c_algorithm_config.A2CAlgorithmConfig, maze.train.trainers.ppo.ppo_algorithm_config.PPOAlgorithmConfig], env: Union[maze.train.parallelization.distributed_env.distributed_env.BaseDistributedEnv, maze.core.env.structured_env.StructuredEnv, maze.core.env.structured_env_spaces_mixin.StructuredEnvSpacesMixin, maze.core.log_stats.log_stats_env.LogStatsEnv], eval_env: [<class 'maze.train.parallelization.distributed_env.distributed_env.BaseDistributedEnv'>, <class 'maze.core.env.structured_env.StructuredEnv'>, <class 'maze.core.env.structured_env_spaces_mixin.StructuredEnvSpacesMixin'>, <class 'maze.core.log_stats.log_stats_env.LogStatsEnv'>], model: maze.core.agent.torch_actor_critic.TorchActorCritic, model_selection: Optional[maze.train.trainers.common.model_selection.best_model_selection.BestModelSelection], initial_state: Optional[str] = None)¶ Base class for multi step actor critic.
- Parameters
algorithm_config – Algorithm parameters.
env – Distributed structured environment
eval_env – Evaluation distributed structured environment
model – Structured torch actor critic model.
initial_state – path to initial state (policy weights, critic weights, optimizer state)
model_selection – Optional model selection class, receives model evaluation results.
-
evaluate(deterministic: bool, repeats: int) → None¶ Perform evaluation on eval env.
- Parameters
deterministic – deterministic or stochastic action sampling (selection)
repeats – number of evaluation episodes to average over