ray.rllib.utils.replay_buffers.multi_agent_prioritized_replay_buffer.MultiAgentPrioritizedReplayBuffer.update_priorities#

MultiAgentPrioritizedReplayBuffer.update_priorities(prio_dict: Dict) → None[source]#

Updates the priorities of underlying replay buffers.

Computes new priorities from td_errors and prioritized_replay_eps. These priorities are used to update underlying replay buffers per policy_id.

Parameters:

  • prio_dict – A dictionary containing td_errors for

  • buffers. (batches saved in underlying replay)

DeveloperAPI: This API may change across minor Ray releases.