mirror of
https://github.com/saymrwulf/stable-baselines3.git
synced 2026-07-13 18:08:39 +00:00
40 lines
1.2 KiB
Markdown
40 lines
1.2 KiB
Markdown
<img src="docs/\_static/img/logo.png" align="right" width="40%"/>
|
|
|
|
[](https://travis-ci.com/hill-a/stable-baselines) [](https://stable-baselines.readthedocs.io/en/master/?badge=master)
|
|
|
|
# Torchy Baselines
|
|
|
|
PyTorch version of [Stable Baselines](https://github.com/hill-a/stable-baselines), a set of improved implementations of reinforcement learning algorithms.
|
|
|
|
## Implemented Algorithms
|
|
|
|
- A2C
|
|
- CEM-RL (with TD3)
|
|
- PPO
|
|
- SAC
|
|
- TD3
|
|
|
|
|
|
## Roadmap
|
|
|
|
TODO:
|
|
- better predict
|
|
- complete logger
|
|
- Refactor: buffer with numpy array instead of pytorch
|
|
- Refactor: remove duplicated code for evaluation
|
|
- double check the shape of log prob
|
|
- try squashing both mean and output when using SAC + SDE
|
|
- plotting? -> zoo
|
|
|
|
Later:
|
|
- get_parameters / set_parameters
|
|
- SDE: use [affine transform](https://www.tensorflow.org/probability/api_docs/python/tfp/bijectors/Affine)
|
|
to scale the noise after a tanh transform?
|
|
- Use MultivariateNormal with full covariance matrix?
|
|
- CNN policies + normalization
|
|
- tensorboard support
|
|
- DQN
|
|
- TRPO
|
|
- ACER
|
|
- DDPG
|
|
- HER -> use stable-baselines because does not depends on tf?
|