search

SforAiDl / genrl

A PyTorch reinforcement learning library for generalizable and reproducible algorithm implementations with an aim to improve accessibility in RL
https://genrl.readthedocs.io
MIT License
403 stars 59 forks source link

Multi Discrete Envs Spaces #295

Closed Sharad24 closed 3 years ago

Sharad24 commented 4 years ago

At multiple places we assume there's only a single dimension in the observation/action space if the env is discrete.

E.g.

https://github.com/SforAiDl/genrl/blob/dc2d58bdade1c5239ccc0925aeda78d1dc59e972/genrl/environments/vec_env/vector_envs.py#L123

https://github.com/SforAiDl/genrl/blob/dc2d58bdade1c5239ccc0925aeda78d1dc59e972/genrl/environments/vec_env/vector_envs.py#L133

sampreet-arthi commented 3 years ago

This caused a bug in running continuous envs with On Policy algos when you tried to fix it ig. For example, PPO wasn't running on Pendulum-v0. I've reverted it back to what it was for now. Let's deal with this when we're working with multi discrete envs?

sampreet-arthi commented 3 years ago

Tracking in #382