Unfortunately, most of the best GitHub repos I’ve found for RL have been with tensorflow. The ones for pytorch that I’ve seen are mediocre at best (in terms of speed, accuracy and code readibility).
What’s the best repo you’ve seen (best in terms of perform using RL algorithms that consume images)? Added bonus if: organized code, prioritized experience replay, and I also like: actor critic, ddpg, dqn agents.
If you have any questions, feel free to ask, open an issue etc.
Not all the methods are implemented yet, but I’m aiming for feature parity with openai baselines, if I have some time I’ll finish implementing ACER this weekend.