ESDevRunner

class maze.train.trainers.es.es_runners.ESDevRunner(state_dict_dump_file: str, spaces_config_dump_file: str, normalization_samples: int, shared_noise_table_size: int, n_eval_rollouts: int)

Runner config for single-threaded training, based on ESDummyDistributedRollouts.

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) → maze.train.trainers.es.distributed.es_distributed_rollouts.ESDistributedRollouts

use single-threaded rollout generation

n_eval_rollouts: int

Fixed number of evaluation runs per epoch.