reinforcement-learning

Why we included this project

This repo is for people who want to understand reinforcement learning by reading actual code, not just theory. Each algorithm lives in one self-contained Python file, so you can go from a paper's update equation to a running example without untangling a big framework. The collection spans classic tabular methods like value iteration and SARSA through deep approaches like DQN, A2C, PPO, and REINFORCE, and it was recently rebuilt on PyTorch with gymnasium and uv, with each file noting the paper it follows and the core equation. The benchmark tables, including Atari runs for DQN and PPO with measured training time and memory on Apple silicon, give you a realistic idea of what these algorithms cost to run. It's a learning resource and reference for students and developers, not something to drop straight into a production system.