stable-baselines3

Stable Baselines3 (SB3) provides a collection of reliable reinforcement learning (RL) algorithm implementations built with PyTorch. It serves as a stable base for the research community and industry to replicate, refine, and compare RL approaches without needing to implement algorithms from scratch. Key features include: - Support for state-of-the-art RL methods with a common, sklearn-like interface. - Support for custom environments, custom policies, and custom callbacks. - Compatibility with Dict observation spaces and Tensorboard for monitoring. - High code coverage, type hints, and PEP8 compliance. - Integration with Weights & Biases for experiment tracking and Hugging Face for model sharing. Implemented algorithms include A2C, ARS, DDPG, DQN, HER, PPO, SAC, TD3, and others. Some experimental features (like Recurrent PPO and TQC) are available via the SB3-Contrib repository, while a Jax-based version (SBX) is available for increased performance. The project also links to the RL Baselines3 Zoo, a training framework that provides scripts for training, evaluating agents, and tuning hyperparameters for common environments.