wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-07-31 12:41:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[rllib] Clean up concepts documentation and policy optimizer creation (#4592)

Browse Source
This commit is contained in:
Eric Liang
2019-04-12 21:03:26 -07:00
committed by GitHub
parent 0f42f87ebc
commit 6e7680bf21
29 changed files with 303 additions and 270 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/rllib/optimizers/sync_replay_optimizer.py
+15 -13
View File
@@ -28,19 +28,21 @@ class SyncReplayOptimizer(PolicyOptimizer):
"td_error" array in the info return of compute_gradients(). This error
term will be used for sample prioritization."""
@override(PolicyOptimizer)
def _init(self,
learning_starts=1000,
buffer_size=10000,
prioritized_replay=True,
prioritized_replay_alpha=0.6,
prioritized_replay_beta=0.4,
schedule_max_timesteps=100000,
beta_annealing_fraction=0.2,
final_prioritized_replay_beta=0.4,
prioritized_replay_eps=1e-6,
train_batch_size=32,
sample_batch_size=4):
def __init__(self,
local_evaluator,
remote_evaluators,
learning_starts=1000,
buffer_size=10000,
prioritized_replay=True,
prioritized_replay_alpha=0.6,
prioritized_replay_beta=0.4,
schedule_max_timesteps=100000,
beta_annealing_fraction=0.2,
final_prioritized_replay_beta=0.4,
prioritized_replay_eps=1e-6,
train_batch_size=32,
sample_batch_size=4):
PolicyOptimizer.__init__(self, local_evaluator, remote_evaluators)
self.replay_starts = learning_starts
# linearly annealing beta used in Rainbow paper