24 Commits

Author SHA1 Message Date
danagi
c59ad40aef
Add auto alpha tuning and exploration noise for sac. (#80)
Add class BaseNoise and GaussianNoise for the concept of exploration noise.
Add new test for sac tested in MountainCarContinuous-v0,
which should benefits from the two above new feature.
2020-06-16 22:17:28 +08:00
Trinkle23897
5f2f05a570 fix #40 2020-06-13 17:06:08 +08:00
Trinkle23897
397e92b0fc fix #77 2020-06-10 12:06:56 +08:00
Trinkle23897
dc451dfe88 nstep all (fix #51) 2020-06-03 13:59:47 +08:00
Trinkle23897
ff81a18f42 compute_nstep_returns (item 2 of #51) 2020-06-02 22:29:50 +08:00
Alexis DUBURCQ
8af7196a9a
Robust conversion from/to numpy/pytorch (#63)
* Enable to convert Batch data back to torch.

* Add torch converter to collector.

* Fix

* Move to_numpy/to_torch convert in dedicated utils.py.

* Use to_numpy/to_torch to convert arrays.

* fix lint

* fix

* Add unit test to check Batch from/to numpy.

* Fix Batch over Batch.

Co-authored-by: Alexis Duburcq <alexis.duburcq@wandercraft.eu>
2020-05-29 20:45:21 +08:00
Trinkle23897
de556fd22d item3 of #51 2020-05-27 11:02:23 +08:00
magicly
6237cc0d52
fix dqn zero eps (#52)
Co-authored-by: liyan <liyan1@digisky.com>
2020-05-21 11:35:41 +08:00
Imone
57bca16f94
Fix log_prob and PPO dual_clip (#49)
* Added DiagGaussian to fix log_probg

* Disable PPO dual_clip
2020-05-18 16:23:35 +08:00
Trinkle23897
0eef0ca198 fix optional type syntax 2020-05-16 20:08:32 +08:00
Trinkle23897
9b26137cd2 add type annotation 2020-05-12 11:31:47 +08:00
Trinkle23897
04b091d975 fix max-grad-norm err in a2c (#46) 2020-05-04 12:33:04 +08:00
Trinkle23897
134f787e24 reserve 'policy' keyword in replay buffer 2020-04-29 17:48:48 +08:00
nicoguertler
8f718d9b13
Fix log_prob in SAC (#41) 2020-04-28 23:44:15 +08:00
Trinkle23897
80d661907e Multimodal obs (#38, #27, #25) 2020-04-28 20:56:02 +08:00
Trinkle23897
959955fa2a fix historical issues 2020-04-26 16:13:51 +08:00
Trinkle23897
6b96f124ae fix pdqn 2020-04-26 15:11:20 +08:00
rocknamx
b23749463e
Prioritized DQN (#30)
* add sum_tree.py

* add prioritized replay buffer

* del sum_tree.py

* fix some format issues

* fix weight_update bug

* simply replace replaybuffer in test_dqn without weight update

* weight default set to 1

* fix sampling bug when buffer is not full

* rename parameter

* fix formula error, add accuracy check

* add PrioritizedDQN test

* add test_pdqn.py

* add update_weight() doc

* add ref of prio dqn in readme.md and index.rst

* restore test_dqn.py, fix args of test_pdqn.py
2020-04-26 12:05:58 +08:00
Trinkle23897
70290346ea compatible with torch==1.5.0 (fix #37) 2020-04-26 11:04:45 +08:00
Trinkle23897
6bf1ea644d fix ppo 2020-04-19 14:30:42 +08:00
Trinkle23897
680fc0ffbe gae 2020-04-14 21:11:06 +08:00
Trinkle23897
3cc22b7c0c __call__ -> forward 2020-04-10 10:47:16 +08:00
Trinkle23897
13086b7f64 add ignore_obs_next in buffer 2020-04-10 09:01:17 +08:00
Trinkle23897
19f2cce294 seealso and change policy dir structure 2020-04-09 21:36:53 +08:00