DistributedActors(env_factory: Callable, Union[maze.core.env.structured_env.StructuredEnv, maze.core.env.structured_env_spaces_mixin.StructuredEnvSpacesMixin, maze.core.log_stats.log_stats_env.LogStatsEnv]], policy: maze.core.agent.torch_policy.TorchPolicy, n_rollout_steps: int, n_actors: int, batch_size: int)¶
The base class for all distributed actors.
Distributed actors run rollouts independently. Rollouts are recorded and made available in batches to be used during training. When a new policy version is made available, it is distributed to all actors.
env_factory – Factory function for envs to run rollouts on.
policy – Structured policy to sample actions from.
n_rollout_steps – Number of rollouts steps to record in one rollout.
n_actors – Number of distributed actors to run simultaneously.
batch_size – Size of the batch the rollouts are collected in.
broadcast_updated_policy(state_dict: Dict) → None¶
Broadcast the newest version of the policy to the actors.
state_dict – State of the new policy version to broadcast.
collect_outputs(learner_device: str) → Tuple[maze.core.trajectory_recording.records.structured_spaces_record.StructuredSpacesRecord, float, float, float]¶
Collect self.batch_size actor outputs from the queue and return them batched where the first dim is time and the second is the batch size.
learner_device – the device of the learner
A tuple of (1) batched version of ActorOutputs, (2) queue size before de-queueing, (3) queue size after dequeueing, and (4) the time it took to dequeue the outputs
get_epoch_stats_aggregator() → maze.core.log_stats.log_stats.LogStatsAggregator¶
Return the collected epoch stats aggregator
get_stats_value(event: Callable, level: maze.core.log_stats.log_stats.LogStatsLevel, name: Optional[str] = None) → Union[int, float, numpy.ndarray, dict]¶
Obtain a single value from the epoch statistics dict.
event – The event interface method of the value in question.
name – The output_name of the statistics in case it has been specified in
level – Must be set to LogStatsLevel.EPOCH, step or episode statistics are not propagated.