opent03 / drl_api

(deep) RL repo with various implementations of DRL algorithms in torch, designed for quick prototyping and implementation of new algorithms.

Deep Reinforcement Learning API (drl_api)

drl_api is a (deep) RL repo with various implementations of DRL algorithms in pytorch. It is designed for quick prototyping and implementation of new algorithms, using its simple and intuitive API.

It is loosely based on Nikolov's rltf.

Maintained regularly, currently supports most Atari NoFrameskip environments and bsuite.

Train model

A highly intuitive example command:

python3 -m run.run_dqn_agent --env_id=BreakoutNoFrameskip-v4 --model=DDQN [--render] [--no-gpu]

Load a saved model and evaluate

Models are saved automatically in drl_api/saves/. A highly intuitive example command:

python3 -m run.play --save_name=BreakoutNoFrameskip-v4-DDQN --rounds=50