all-rl-algorithms

Why we included this project

This repository is a set of eighteen Jupyter notebooks that implement reinforcement learning algorithms from scratch, using only NumPy, Matplotlib, and PyTorch. Each notebook covers one algorithm, from tabular Q-Learning and SARSA up to PPO, SAC, DDPG, and multi-agent approaches like MADDPG and QMIX, with code that stays close to the underlying math. The author intentionally favors readability over speed, so the result reads like an interactive textbook rather than a production library. A cheat sheet collects the formulas and pseudocode, and since everything runs in notebooks, you can adjust hyperparameters and immediately see how behavior changes. That makes it a practical resource for students and researchers, or engineers who want to understand the mechanics behind value-based and policy-gradient methods, as well as model-based ones, before moving to heavier frameworks.