Closed eugenevinitsky closed 5 years ago
Relatedly, we can integrate with https://github.com/hill-a/stable-baselines.
I do really like the idea of supporting a library with lightweight implementations of RL methods.
Another library to consider is garage (https://github.com/rlworkgroup/garage) based on rllab. See also #637 .
Yeah, since we've already built all this support for rllab garage makes sense!
https://github.com/openai/baselines is a pretty common starting point for RL training. It'd be neat to integrate it. Alternately, we could also integrate Google's Dopamine: https://github.com/google/dopamine
The process would be: (1) Add an example script that successfully uses one of these frameworks. (2) Add a link to how to use it to the setup instructions.