wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-07-27 11:26:41 +08:00
Code Issues Packages Projects Releases Wiki Activity

[rllib] Fix edge case in n-step calculation and non-apex replay prioritization (#2929)

Browse Source 
* fix

* lint
This commit is contained in:
Eric Liang
2018-09-28 15:22:33 -07:00
committed by GitHub
parent 4ffe1e3556
commit f1c55497ce
3 changed files with 23 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/rllib/optimizers/sync_replay_optimizer.py
+5 -5
View File
@@ -92,13 +92,13 @@ class SyncReplayOptimizer(PolicyOptimizer):
for policy_id, s in batch.policy_batches.items():
for row in s.rows():
if "weights" not in row:
row["weights"] = np.ones_like(row["rewards"])
self.replay_buffers[policy_id].add(
pack_if_needed(row["obs"]),
row["actions"], row["rewards"],
pack_if_needed(row["new_obs"]), row["dones"],
row["weights"])
row["actions"],
row["rewards"],
pack_if_needed(row["new_obs"]),
row["dones"],
weight=None)
if self.num_steps_sampled >= self.replay_starts:
self._optimize()