This repository provides educational, from-scratch Python implementations of 18 Reinforcement Learning (RL) algorithms, targeting students and researchers seeking a deep, intuitive understanding of RL mechanics. It prioritizes code clarity and pedagogical value over performance optimization, serving as an interactive textbook with accompanying Jupyter Notebooks and a comprehensive RL cheat sheet.

How It Works

The project implements RL algorithms using fundamental libraries like NumPy, Matplotlib, and PyTorch, avoiding complex abstractions. Each algorithm is presented in a dedicated Jupyter Notebook, detailing its core logic, mathematical underpinnings, and providing runnable code for experimentation. This approach allows users to directly modify parameters and observe effects, fostering hands-on learning.

Quick Start & Requirements

• Install: pip install -r requirements.txt after cloning the repository.
• Prerequisites: Python 3.x, NumPy, Matplotlib, PyTorch.
• Usage: Run algorithms via Jupyter Notebooks (.ipynb files). A3C implementation requires running a3c_training.py from the terminal.
• Resources: Virtual environment recommended. Setup time is minimal, dependent on dependency installation.
• Docs: Repository Link

Highlighted Details

• Implements 18 RL algorithms including Q-Learning, SARSA, REINFORCE, PPO, DQN, SAC, MADDPG, and PlaNet.
• Features a comprehensive RL Cheat Sheet summarizing core concepts, algorithms, and hyperparameters.
• Jupyter Notebooks offer step-by-step explanations and visualizations of learning curves and policies.
• Code is designed for readability and clarity, closely mirroring mathematical descriptions.

Maintenance & Community

The repository is actively updated, with recent additions including 18 new algorithms and a cheat sheet. Contributions are welcomed via pull requests for bug fixes, improved explanations, or new algorithm implementations. Users can create issues to discuss contributions.

Licensing & Compatibility

The repository's license is not explicitly stated in the provided README. Users should verify licensing for commercial or closed-source integration.

Limitations & Caveats

This repository is primarily for educational purposes and is not performance-optimized. Some implementations, particularly for more complex algorithms, may contain bugs, simplifications, or be incomplete.