Pearl
Basic Information
Pearl is an open-source, production-ready reinforcement learning agent library released by the Applied Reinforcement Learning team at Meta to help researchers and practitioners build customizable RL agents. It is designed for developing agents that prioritize cumulative long-term feedback and that can operate under limited observability, sparse feedback, and high stochasticity. The repository provides modular building blocks—policy learners, exploration modules, replay buffers, environment wrappers and action representation modules—so users can assemble agents for research experiments or real-world systems. The README includes installation instructions (pip install -e .), a quickstart example using a Gym environment, multiple tutorial notebooks (recommender, contextual bandits, Frozen Lake, DQN/DDQN, actor-critic with safety), and notes on adoption in recommender systems, auction bidding, and creative selection.
Links
App Details
-
$$$$ -
$$$$