stable-baselines3

Train reinforcement learning agents using Stable Baselines3 with a unified Python API supporting PPO, SAC, TD3, and DQN across diverse environments. It provides vectorized training, evaluation utilities, and model persistence for research and production workflows.

Vectorized environments and callbacks are supported to accelerate reinforcement learning training. The Skill enables parallel environment execution alongside custom training callbacks, evaluation utilities, and TensorBoard logging for experiment tracking.

Gymnasium and Stable Baselines3 must be installed in your Python environment to run reinforcement learning experiments. These dependencies provide the foundational environment interfaces and algorithm implementations required for training and deploying agents.

The best way to start with reinforcement learning for discrete and continuous control tasks is training a PPO agent on CartPole-v1. This establishes a baseline using the Skill's unified API, allowing rapid prototyping and iteration for more complex environments.

Stable Baselines3 supports model persistence and optional VecNormalize for deploying trained RL agents. You can save and load trained models, normalize vectorized environment observations, and utilize TensorBoard logging to monitor deployment-ready experiments.

Use Stable Baselines3 over other reinforcement learning libraries for production experiments due to its stable, well-documented algorithms and consistent Python interface. It accelerates workflows with comprehensive training utilities, vectorized environments, and model persistence across PPO, SAC, TD3, and DQN.