ESLocalRunner

class maze.train.trainers.es.es_runners.ESLocalRunner(state_dict_dump_file: str, dump_interval: Optional[int], spaces_config_dump_file: str, normalization_samples: int, shared_noise_table_size: int, n_train_workers: int, n_eval_workers: int, start_method: str)

Runner config for multi-process training, based on ESSubprocDistributedRollouts.

create_distributed_rollouts(env: Union[maze.core.env.structured_env.StructuredEnv, maze.core.env.structured_env_spaces_mixin.StructuredEnvSpacesMixin], shared_noise: maze.train.trainers.es.es_shared_noise_table.SharedNoiseTable, agent_instance_seed: int) → maze.train.trainers.es.distributed.es_distributed_rollouts.ESDistributedRollouts

(overrides ESMasterRunner)

use multi-process rollout generation

n_eval_workers: int

Number of worker processes to spawn for evaluation

n_train_workers: int

Number of worker processes to spawn for training

start_method: str

Type of start method used for multiprocessing (‘forkserver’, ‘spawn’, ‘fork’).