Reinforcement Learning Kitchen Sink

PyTorch Implementations of Standard Deep RL Algorithms (including REINFORCE, A2C, PPO). Each of the below algorithms are implemented for both OpenAI Gym Classic Control Tasks (e.g. Cartpole, MountainCar), as well as the Atari suite.

This repository exists mostly as a means to illustrate basic RL algorithms via extremely readable and well-documented PyTorch implementations. This repository also stores hyperparameters and learning curves (with confidence intervals!) for each task.

This code is inspired in equal parts from the OpenAI Baselines and Ilya Kostrikov's PyTorch-RL Repository.

REINFORCE

Advantage Actor-Critic (A2C)

Proximal Policy Optimization