ReAgent
An end-to-end platform for applied reinforcement learning and contextual bandits, originally developed at Facebook for production recommendation systems.
What is ReAgent?
ReAgent is an open-source platform for applied reinforcement learning and decision-making systems, originally developed and used at Facebook. It provides an end-to-end solution for training, evaluating, and serving RL models in production environments, particularly for recommendation systems and optimization tasks where simulators aren't available.
Machine learning engineers and researchers building production reinforcement learning systems, particularly those working on recommendation systems, contextual bandits, or batch RL applications.
Developers choose ReAgent for its comprehensive production-ready workflow, support for safety-focused tools like counterfactual policy evaluation, and its origin as Facebook's internal RL platform designed for real-world applications.
Overview
A platform for Reasoning systems (Reinforcement Learning, Contextual Bandits, etc.)
Use Cases
Best For
- Building production reinforcement learning systems for recommendation engines
- Implementing contextual bandit algorithms for dynamic decision-making
- Training offline/batch RL models when simulators are unavailable
- Evaluating policies safely using counterfactual evaluation methods
- Developing slate recommendation systems with algorithms like Seq2Slate
- Deploying RL models with PyTorch and TorchScript serving
Not Ideal For
- Projects requiring real-time, online reinforcement learning with frequent policy updates
- Academic research focused on experimenting with the latest RL algorithms not included in the library
- Small teams or startups needing a lightweight, easy-to-setup RL solution for rapid prototyping
- Any new project due to its archived status and lack of ongoing maintenance
Pros & Cons
Pros
Provides a complete pipeline from data preprocessing to model serving with TorchScript, as described in the overview for large-scale, distributed RL tasks.
Includes counterfactual policy evaluation methods like Doubly Robust and MAGIC, along with behavior cloning, to safely bootstrap and assess policies without deployment risks.
Offers specialized algorithms such as Seq2Slate and SlateQ, tailored for slate recommendation tasks common in real-world applications like those at Facebook.
Developed and used internally at Facebook, ensuring reliability and practicality for applied RL in production environments, as noted in the release post.
Cons
The project is officially archived with no future updates or support, and users are directed to Pearl for ongoing development, limiting long-term viability.
Designed for large-scale, distributed tasks, installation via Docker or manual processes is cumbersome, making it overkill for smaller or simpler projects.
Built exclusively on PyTorch and TorchScript, it doesn't support TensorFlow or other ML frameworks, restricting flexibility for diverse tech stacks.
Emphasizes offline training and slow policy updates, so it lacks support for real-time online RL scenarios, as stated in the tutorial.
Frequently Asked Questions
Related Projects
rayRay is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads.
gymA toolkit for developing and comparing reinforcement learning algorithms.
fastaiThe fastai deep learning library
mlflowThe open source AI engineering platform for agents, LLMs, and ML models. MLflow enables teams of all sizes to debug, evaluate, monitor, and optimize production-quality AI applications while controlling costs and managing access to models and data.
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.