ESDummyDistributedRollouts

class maze.train.trainers.es.distributed.es_dummy_distributed_rollouts.ESDummyDistributedRollouts(env: StructuredEnv, n_eval_rollouts: int, shared_noise: SharedNoiseTable, agent_instance_seed: int)

Implementation of the ES distribution by running the rollouts synchronously in the same process.

generate_rollouts(policy: Policy | TorchModel, max_steps: int | None, noise_stddev: float, normalization_stats: Dict[str, Dict[str, numpy.ndarray | float | int | Iterable[float | int]]]) → Generator[ESRolloutResult, None, None]

(overrides ESDistributedRollouts)

First execute a fixed number of eval rollouts and then continue with producing training samples.