34bae27ac7d3c39335a2b8c7d4cd4efba56aa398,rllib/execution/replay_buffer.py,LocalReplayBuffer,add_batch,#LocalReplayBuffer#Any#,348

Before Change

After Change

        // Handle everything as if multiagent
        if isinstance(batch, SampleBatch):
            batch = MultiAgentBatch({DEFAULT_POLICY_ID: batch}, batch.count)
        with self.add_batch_timer:
            if self.replay_mode == "lockstep":
                // Note that prioritization is not supported in this mode.
                for s in batch.timeslices(self.replay_sequence_length):
                    self.replay_buffers[_ALL_POLICIES].add(s, weight=None)
            else:
                for policy_id, b in batch.policy_batches.items():
                    for s in b.timeslices(self.replay_sequence_length):
                        if "weights" in s:
                            weight = np.mean(s["weights"])
                        else:
                            weight = None
                        self.replay_buffers[policy_id].add(s, weight=weight)
        self.num_added += batch.count

    def replay(self):
        if self._fake_batch:

Instances