stable-baselines3

Train a reinforcement learning agent with PPO by setting up a Gymnasium-compatible environment, selecting the PPO algorithm from Stable Baselines3, and calling the unified train API to begin optimization. This Skill provides templates that streamline the entire training workflow from environment initialization through model persistence.

Stable Baselines3 supports PPO, SAC, TD3, DQN, and A2C algorithms for reinforcement learning, all accessible through a unified API compatible with Gymnasium environments. Each algorithm implementation is production-ready and supports vectorized environments for scalable experimentation.

Vectorized environments enable parallel reinforcement learning training by running multiple environment instances simultaneously through Stable Baselines3, significantly improving sample throughput and training efficiency. This Skill provides guidance and templates for configuring both parallel and single-agent setups.

Evaluate a trained RL agent across Gymnasium environments using Stable Baselines3's built-in evaluation workflows, which provide standardized performance metrics and support callbacks for monitoring during evaluation. The Skill includes templates for evaluation and experimentation to ensure consistent assessment.

Gymnasium is a required dependency for Stable Baselines3 algorithms, as the framework enforces Gymnasium compatibility for all environment interfaces and task definitions. You also need NumPy installed for array operations supporting training and evaluation workflows.

Save and load trained reinforcement learning models using Stable Baselines3's model persistence functionality, which allows storing trained agents and reloading them for continued training or deployment. This Skill covers model persistence alongside callbacks and monitoring for complete experiment management.