Welcome to Lightning-Baselines3 docs - Reliable and Modular Reinforcement Learning Implementations

Lightning Baselines3 (LB3) is a set of reliable and modular implementations of reinforcement learning algorithms built on top of PyTorch Lightning. It is adapted from Stable Baselines3 (SB3)

Github repository: https://github.com/HenryJia/lightning-baselines3

Main Features

Implemented

• Unified structure for all algorithms

• Tests, high code coverage and type hints

• Modular, flexible and clean code in line with the Lightning Philosophy

• Documented functions and classes

Coming Soon

• PEP8 compliant (unified code style)

• The performance of each algorithm was tested (see Results section in their respective page)

User Guide

• Installation
  • Prerequisites
  • Development version
• Lightning Baselines3 in 3 Steps
  • Step 1: Choose an algorithm
  • Step 2: Define Your Model
  • Step 2: Fit with Lightning Trainer
• Reinforcement Learning Tips and Tricks
  • General advice when using Reinforcement Learning
  • Which algorithm should I use?
  • Tips and Tricks when creating a custom environment
  • Tips and Tricks when implementing an RL algorithm
• Reinforcement Learning Resources
• RL Algorithms
  • Reproducibility
• Using Custom Environments

RL Algorithms

• Base RL Class
• A2C
• PPO
• DQN
• TD3
• DDPG
• SAC

Citing Stable Baselines3

To cite this project in publications:

@misc{lightning-baselines3,
  author = {Jia, Hengjian},
  title = {Lightning Baselines3},
  year = {2021},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/HenryJia/lightning-baselines3}},
}

Indices and tables

• Index

• Search Page

• Module Index