wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-09-10 12:38:43 +08:00
Code Issues Packages Projects Releases Wiki Activity

[rllib] clarify train batch size for PPO (#2793)

Browse Source 
It's possible to configure PPO in a way that ends up discarding most of the samples (they are treated as "stragglers"). Add a warning when this happens, and raise an exception if the waste is particularly egregious.
This commit is contained in:
Eric Liang
2018-09-05 12:06:13 -07:00
committed by GitHub
parent c87a9114cd
commit 995ac24a2c
24 changed files with 83 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/rllib/test/test_supported_spaces.py
+4 -3
View File
@@ -107,15 +107,16 @@ class ModelSupportedSpaces(unittest.TestCase):
"PPO", {
"num_workers": 1,
"num_sgd_iter": 1,
"timesteps_per_batch": 1,
"sgd_batchsize": 1
"train_batch_size": 10,
"sample_batch_size": 10,
"sgd_minibatch_size": 1
}, stats)
check_support(
"ES", {
"num_workers": 1,
"noise_size": 10000000,
"episodes_per_batch": 1,
"timesteps_per_batch": 1
"train_batch_size": 1
}, stats)
check_support(
"ARS", {