hongshaorou/Tianshou
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
288 Commits 3 Branches 28 Tags
Commit Graph

233 Commits

Author SHA1 Message Date
Jiayi Weng
b9eedc516e bump to 0.4.4 2021-10-13 12:22:24 -04:00
Jiayi Weng
926ec0b9b1
update save_fn in trainer (#459) 
- collector.collect() now returns 4 extra keys: rew/rew_std/len/len_std (previously this work is done in logger)
- save_fn() will be called at the beginning of trainer
 2021-10-13 21:25:24 +08:00
Jiayi Weng
e45e2096d8
add multi-GPU support (#461) 
add a new class DataParallelNet
 2021-10-06 01:39:14 +08:00
Jiayi Weng
5df64800f4
final fix for actor_critic shared head parameters (#458) 2021-10-04 23:19:07 +08:00
Ayush Chaurasia
22d7bf38c8
Improve W&B logger (#441) 
- rename WandBLogger -> WandbLogger
- add save_data and restore_data
- allow more input arguments for wandb init
- integrate wandb into test/modelbase/test_psrl.py and examples/atari/atari_dqn.py
- documentation update
 2021-09-24 21:52:23 +08:00
n+e
fc251ab0b8
bump to v0.4.3 (#432) 
* add makefile
* bump version
* add isort and yapf
* update contributing.md
* update PR template
* spelling check
 2021-09-03 05:05:04 +08:00
Ending Hsiao
a740496a51
fix dual clip implementation (#435) 
close #433
 2021-09-02 21:43:14 +08:00
Andriy Drozdyuk
8a5e2190f7
Add Weights and Biases Logger (#427) 
- rename BasicLogger to TensorboardLogger
- refactor logger code
- add WandbLogger

Co-authored-by: Jiayi Weng <trinkle23897@gmail.com>
 2021-08-30 22:35:02 +08:00
n+e
e4f4f0e144
fix docs build failure and a bug in a2c/ppo optimizer (#428) 
* fix rtfd build

* list + list -> set.union

* change seed of test_qrdqn

* add py39 test
 2021-08-30 02:07:03 +08:00
Yi Su
291be08d43
Add Rainbow DQN (#386) 
- add RainbowPolicy
- add `set_beta` method in prio_buffer
- add NoisyLinear in utils/network
 2021-08-29 23:34:59 +08:00
Andriy Drozdyuk
d161059c3d
Replaced indice by plural indices (#422) 2021-08-20 21:58:44 +08:00
Andriy Drozdyuk
18d2f25eff
Remove warnings about the use of save_fn across trainers (#408) 2021-08-04 09:56:00 +08:00
n+e
c19876179a
add env_id in preprocess fn (#391) 2021-07-05 09:50:39 +08:00
n+e
ebaca6f8da
add vizdoom example, bump version to 0.4.2 (#384) 2021-06-26 18:08:41 +08:00
Yi Su
c0bc8e00ca
Add Fully-parameterized Quantile Function (#376) 2021-06-15 11:59:02 +08:00
Yi Su
f3169b4c1f
Add Implicit Quantile Network (#371) 2021-05-29 09:44:23 +08:00
Ark
655d5fb14f
Allow researchers to choose whether to use Double DQN (#368) 2021-05-21 10:53:34 +08:00
Yi Su
8f7bc65ac7
Add discrete Critic Regularized Regression (#367) 2021-05-19 13:29:56 +08:00
Yi Su
b5c3ddabfa
Add discrete Conservative Q-Learning for offline RL (#359) 
Co-authored-by: Yi Su <yi.su@antgroup.com>
Co-authored-by: Yi Su <yi.su@antfin.com>
 2021-05-12 09:24:48 +08:00
Ark
84f58636eb
Make trainer resumable (#350) 
- specify tensorboard >= 2.5.0
- add `save_checkpoint_fn` and `resume_from_log` in trainer

Co-authored-by: Trinkle23897 <trinkle23897@gmail.com>
 2021-05-06 08:53:53 +08:00
Yuge Zhang
f4e05d585a
Support deterministic evaluation for onpolicy algorithms (#354) 2021-04-27 21:22:39 +08:00
n+e
ff4d3cd714
Support different state size and fix exception in venv.__del__ (#352) 
- Batch: do not raise error when it finds list of np.array with different shape[0].

- Venv's obs: add try...except block for np.stack(obs_list)

- remove venv.__del__ since it is buggy
 2021-04-25 15:23:46 +08:00
ChenDRAG
844d7703c3
NPG Mujoco benchmark release (#347) 2021-04-21 16:31:20 +08:00
ChenDRAG
1dcf65fe21
Add NPG policy (#344) 2021-04-21 09:52:15 +08:00
n+e
c059f98abf
fix atari_bcq (#345) 2021-04-20 22:59:21 +08:00
ChenDRAG
a57503c0aa
TRPO benchmark release (#340) 2021-04-19 17:05:06 +08:00
ChenDRAG
5057b5c89e
Add TRPO policy (#337) 2021-04-16 20:37:12 +08:00
ChenDRAG
dd4a01132c
Fix SAC loss explode (#333) 
* change SAC action_bound_method to "clip" (tanh is hardcoded in forward)

* docstring update

* modelbase -> modelbased
 2021-04-04 17:33:35 +08:00
n+e
825da9bc53
add cross-platform test and release 0.4.1 (#331) 
* bump to 0.4.1

* add cross-platform test
 2021-03-31 15:14:22 +08:00
n+e
09692c84fe
fix numpy>=1.20 typing check (#323) 
Change the behavior of to_numpy and to_torch: from now on, dict is automatically converted to Batch and list is automatically converted to np.ndarray (if an error occurs, raise the exception instead of converting each element in the list).
 2021-03-30 16:06:03 +08:00
ChenDRAG
6426a39796
ppo benchmark (#330) 2021-03-30 11:50:35 +08:00
ChenDRAG
5d580c3662
refactor ppo (#329) 2021-03-28 18:28:36 +08:00
ChenDRAG
1730a9008a
A2C benchmark for mujoco (#325) 2021-03-28 13:12:43 +08:00
ChenDRAG
105b277b87
hotfix:keep statisics of buffer when reset buffer in on policy trainer (#328) 2021-03-27 16:58:48 +08:00
ChenDRAG
3ac67d9974
refactor A2C/PPO, change behavior of value normalization (#321) 2021-03-25 10:12:39 +08:00
ChenDRAG
e27b5a26f3
Refactor PG algorithm and change behavior of compute_episodic_return (#319) 
- simplify code
- apply value normalization (global) and adv norm (per-batch) in on-policy algorithms
 2021-03-23 22:05:48 +08:00
ChenDRAG
2c11b6e43b
Add lr_scheduler option for Onpolicy algorithm (#318) 
add lr_scheduler option in PGPolicy/A2CPolicy/PPOPolicy
 2021-03-22 16:57:24 +08:00
ChenDRAG
4d92952a7b
Remap action to fit gym's action space (#313) 
Co-authored-by: Trinkle23897 <trinkle23897@gmail.com>
 2021-03-21 16:45:50 +08:00
n+e
ec23c7efe9
fix qvalue mask_action error for obs_next (#310) 
* fix #309
* remove for-loop in dqn expl_noise
 2021-03-15 08:06:24 +08:00
ChenDRAG
243ab43b3c
support observation normalization in BaseVectorEnv (#308) 
add RunningMeanStd
 2021-03-11 20:50:20 +08:00
ChenDRAG
5c53f8c1f8
fix reward_metric & n_episode bug in on policy algorithm (#306) 2021-03-08 14:35:30 +08:00
ChenDRAG
e605bdea94
MuJoCo Benchmark - DDPG, TD3, SAC (#305) 
Releasing Tianshou's SOTA benchmark of 9 out of 13 environments from the MuJoCo Gym task suite.
 2021-03-07 19:21:02 +08:00
n+e
454c86c469
fix venv seed, add TOC in docs, and split buffer.py into several files (#303) 
Things changed in this PR:

- various docs update, add TOC
- split buffer into several files
- fix venv action_space randomness
 2021-03-02 12:28:28 +08:00
n+e
31e7f445d1
fix vecenv action_space randomness (#300) 2021-03-01 15:44:03 +08:00
ChenDRAG
f22b539761
Remove reward_normaliztion option in offpolicy algorithm (#298) 
* remove rew_norm in nstep implementation
* improve test
* remove runnable/
* various doc fix

Co-authored-by: n+e <trinkle23897@gmail.com>
 2021-02-27 11:20:43 +08:00
ChenDRAG
3108b9db0d
Add Timelimit trick to optimize policies (#296) 
* consider timelimit.truncated in calculating returns by default
* remove ignore_done
 2021-02-26 13:23:18 +08:00
ChenDRAG
9b61bc620c add logger (#295) 
This PR focus on refactor of logging method to solve bug of nan reward and log interval. After these two pr, hopefully fundamental change of tianshou/data is finished. We then can concentrate on building benchmarks of tianshou finally.

Things changed:

1. trainer now accepts logger (BasicLogger or LazyLogger) instead of writer;
2. remove utils.SummaryWriter;
 2021-02-24 14:48:42 +08:00
Trinkle23897
e99e1b0fdd Improve buffer.prev() & buffer.next() (#294) 2021-02-22 19:19:22 +08:00
ChenDRAG
7036073649
Trainer refactor : some definition change (#293) 
This PR focus on some definition change of trainer to make it more friendly to use and be consistent with typical usage in research papers, typically change `collect-per-step` to `step-per-collect`, add `update-per-step` / `episode-per-collect` accordingly, and modify the documentation.
 2021-02-21 13:06:02 +08:00
ChenDRAG
150d0ec51b
Step collector implementation (#280) 
This is the third PR of 6 commits mentioned in #274, which features refactor of Collector to fix #245. You can check #274 for more detail.

Things changed in this PR:

1. refactor collector to be more cleaner, split AsyncCollector to support asyncvenv;
2. change buffer.add api to add(batch, bffer_ids); add several types of buffer (VectorReplayBuffer, PrioritizedVectorReplayBuffer, etc.)
3. add policy.exploration_noise(act, batch) -> act
4. small change in BasePolicy.compute_*_returns
5. move reward_metric from collector to trainer
6. fix np.asanyarray issue (different version's numpy will result in different output)
7. flake8 maxlength=88
8. polish docs and fix test

Co-authored-by: n+e <trinkle23897@gmail.com>
 2021-02-19 10:33:49 +08:00