wassname/DeepRL
1
0
Fork 0
mirror of https://github.com/wassname/DeepRL.git synced 2026-08-20 12:00:17 +08:00
Code Issues Packages Projects Releases Wiki Activity
124 Commits 11 Branches 0 Tags
master
Commit Graph
124 Commits
This Branch
This Branch
All Branches
Author SHA1 Message Date
wassname aeae2c5d58 Merge branch 'master' of https://github.com/ShangtongZhang/DeepRL 2018-01-15 14:05:32 +08:00
Shangtong Zhang 8224a41ba9 Update README 2018-01-01 22:49:19 -07:00
Shangtong Zhang b85ec446aa Introduce tensorboardX 2018-01-01 21:52:24 -07:00
Shangtong Zhang 71fa1aaf48 Minor fix 2018-01-01 20:45:56 -07:00
Shangtong Zhang 71002e4109 A single thread DDPG 2018-01-01 20:45:15 -07:00
Shangtong Zhang 0b075a353b Reseed for each worker 2017-12-25 09:52:38 -07:00
Shangtong Zhang 964734755f Expose original Atari game reward 2017-12-23 20:52:17 -07:00
Shangtong Zhang 69b7ea53cc Minor update for DQN 2017-12-21 22:22:36 -07:00
Shangtong Zhang 8839614c44 Minor fix for python3 2017-12-20 11:40:06 -07:00
Shangtong Zhang 7a97c742cf Update README 2017-12-20 09:40:03 -07:00
Shangtong Zhang 1b9acb17f0 Complete ACVP 2017-12-20 09:36:37 -07:00
Shangtong Zhang 040df4efcf Print pictures 2017-12-19 22:09:35 -07:00
Shangtong Zhang 13533ab2fe Set hidden size 2017-12-18 20:09:11 -07:00
Shangtong Zhang e4c7558726 Update video prediction 2017-12-18 09:28:09 -07:00
Shangtong Zhang f0940e86dc Setup 1-step prediction 2017-12-16 22:14:40 -07:00
Shangtong Zhang 214f7693c2 Generate dataset for video prediction 2017-12-16 20:03:11 -07:00
Shangtong Zhang 56304e63e6 Update logger 2017-12-14 15:12:36 -07:00
Shangtong Zhang fed4bd5163 Update async methods 2017-12-12 10:39:16 -07:00
Shangtong Zhang 83a4270bb2 Upgrade to PyTorch v0.3 2017-12-11 23:41:16 -07:00
Mike Clark 034e2dfe09 replace flip function 2017-11-21 08:40:23 +08:00
Mike Clark 574dbd6ecc vectorised gae calc 
vectorised advantage calculation turns out to be 40x faster, modified from here https://github.com/pmlg/deep-rl-bootcamp/blob/master/lab4/alg_utils.py#L35
 2017-11-21 08:37:33 +08:00
Shangtong Zhang 4697d1d3e9 Remove unnecessary conversion 2017-11-14 19:14:42 -07:00
Shangtong Zhang 65eea71bdb Merge branch 'wassname-master' 2017-11-14 19:12:44 -07:00
wassname e16a560abc let ppo work with gpu 2017-11-15 09:40:58 +08:00
wassname aec67c66f8 Merge branch 'master' of https://github.com/ShangtongZhang/DeepRL 2017-11-15 09:38:53 +08:00
Shangtong Zhang d47bb7a59a Update README 2017-11-11 21:05:03 -07:00
Shangtong Zhang 0fb70fa484 Minor fix 2017-11-11 20:58:18 -07:00
Shangtong Zhang c1fdb1bd2f Major update 2017-11-11 20:52:57 -07:00
Shangtong Zhang 8c197e183c Support D3PG 2017-11-11 19:41:52 -07:00
Shangtong Zhang edc4e89b9b Update DDPG 2017-11-11 15:27:06 -07:00
Shangtong Zhang 41627588fb Action dim 2017-11-09 16:38:42 -07:00
Shangtong Zhang e1e6446728 Minor update 2017-11-09 08:50:57 -07:00
Shangtong Zhang 347d0d22cb Fix a bug 2017-11-05 18:58:14 -07:00
Shangtong Zhang 34c6ca7dd8 Update figures 2017-11-05 00:07:38 -06:00
Shangtong Zhang 29c6ce7266 Update figures 2017-11-05 00:07:07 -06:00
Shangtong Zhang 736825a383 Update README 2017-11-05 00:02:15 -06:00
Shangtong Zhang cc1e1415c7 Merge pull request #5 from wassname/patch-2 
add eps to log_density to avoid NaN
 2017-11-04 09:13:29 -06:00
Mike Clark deab7043fc add eps to avoid NaN 
I tracked some NaN's I was getting down to here. It happens when std is a small number, then var is even smaller, and log_density=inf. There are some problems where the agent will learn to use small standard deviations because jittering movements have a high cost, this makes those more stable.
 2017-11-04 16:24:54 +08:00
wassname 662c0834f6 This helps me avoid NaN when I have small rewards 2017-11-01 18:58:03 +08:00
Shangtong Zhang 9ef309696f Merge pull request #4 from wassname/patch-1 
moving epsilon outside softplus
 2017-10-31 21:01:10 -06:00
Mike Clark 212ff0bc70 moving epsilon outside softplus 
I think you intended this to be outside the softplus. The reason is that it should be applied just before the log to avoid `log(0)=inf`.e.g.

- `log(softplus(-1000+1e-5))=log(0)=inf`. 
- `log(softplus(-1000)+1e-5)=log(1e-5)!=inf`. 

Also this fixes a NaN I had.
 2017-11-01 10:52:23 +08:00
Shangtong Zhang 55ae92619a Update README 2017-10-28 11:08:25 -06:00
Shangtong Zhang 9615195f5e Upgrade to PyTorch v0.2.0 2017-10-28 11:06:20 -06:00
Shangtong Zhang a59f2d2911 Benchmark DDPG with Roboschool 2017-10-27 23:09:24 -06:00
Shangtong Zhang af360bb9ea Support roboschool 2017-10-27 14:23:54 -06:00
Shangtong Zhang e832e2e89a Update README 2017-10-26 23:06:20 -06:00
Shangtong Zhang 6682937b54 Support python3 and pytorch 0.2 2017-10-26 23:01:57 -06:00
Shangtong Zhang 1b6de16b4f Fix a critical bug in DDPG 2017-10-15 10:05:33 -06:00
Shangtong Zhang 3ed6f2a9db Log wall time 2017-10-14 16:38:51 -06:00
Shangtong Zhang c04b2a395f Update README 2017-10-10 23:02:28 -06:00