search

coax-dev / coax

Modular framework for Reinforcement Learning in python
https://coax.readthedocs.io
MIT License
168 stars 17 forks source link

Add dm_control example for SAC #26

Closed frederikschubert closed 2 years ago

frederikschubert commented 2 years ago

This PR introduces the common squashed normal distribution for the SAC policy on dm_control and provides an example that solves the walker.walk task. Interestingly clipping the actions to the range [-1, 1] diverges. rendering

@KristianHolsheimer How would you go about changing the installation script for this notebook to add dm_control as a dependency?

KristianHolsheimer commented 2 years ago

Perhaps it would be good to add install_extras={'box2d': [...], 'atari': [...], 'dm_control': [...]} in setup.py to avoid depending on all those envs by default. You would then have to install using e.g.

$ pip install coax[dm_control]

But for now I would just add it to requirements.txt. If it becomes an issue we can always separate them out later.

frederikschubert commented 2 years ago

Sounds good 👍 I will update the documentation tomorrow so that this example is rendered in the correct section.

KristianHolsheimer commented 2 years ago

Awesome, thanks!

frederikschubert commented 2 years ago

I will close this PR and open a new one that builds on the two update PRs https://github.com/coax-dev/coax/pull/27 and https://github.com/coax-dev/coax/pull/28.