ray.rllib.utils.replay_buffers.prioritized_replay_buffer.PrioritizedReplayBuffer#

class ray.rllib.utils.replay_buffers.prioritized_replay_buffer.PrioritizedReplayBuffer(capacity: int = 10000, storage_unit: str = 'timesteps', alpha: float = 1.0, **kwargs)[source]#

Bases: ray.rllib.utils.replay_buffers.replay_buffer.ReplayBuffer

This buffer implements Prioritized Experience Replay.

The algorithm has been described by Tom Schaul et. al. in “Prioritized Experience Replay”. See https://arxiv.org/pdf/1511.05952.pdf for the full paper.

DeveloperAPI: This API may change across minor Ray releases.

Methods

__init__([capacity, storage_unit, alpha])

Initializes a PrioritizedReplayBuffer instance.

add(batch, **kwargs)

Adds a batch of experiences or other data to this buffer.

apply(func, *args, **kwargs)

Calls the given function with this rollout worker instance.

get_host()

Returns the computer's network name.

get_state()

Returns all local state.

ping()

Ping the actor.

sample(num_items, beta, **kwargs)

Sample num_items items from this buffer, including prio.

set_state(state)

Restores all local state to the provided state.

stats([debug])

Returns the stats of this buffer.

update_priorities(idxes, priorities)

Update priorities of items at given indices.