TRAINING: Introduction to reinforcement learning and RLlib

Reinforcement learning (RL) is gaining traction as a complementary approach to supervised learning for applications from recommender systems to games to production planning. This has been partly due to the ability of new simulators to run code in parallel, thus speeding up training times.

This course will teach you how to create a multiplayer game, solvable by RL, using RLlib. You will learn the basics of RL and how to use RLlib to train the different behaviors of the game's characters. We will then optimize our models' hyperparameters with Ray Tune and also learn how to add an in-game item recommendation system using offline RL.

Using Jupyter notebooks running on Anyscale in the cloud, we will combine theoretical concepts of RL with practical exercises. You will leave with a complete, working example of parallelized Python RL code using Ray RLlib on a GitHub repo.

Key takeaways:

• Learn key deep RL concepts and terminology
• Get an overview of the state-of-the-art RL algorithms available in RLlib
• Learn how to customize an RL environment using OpenAI Gym
• Train and tune an RLlib algorithm (SlateQ and multi-agent PPO) using Ray Tune
• Checkpoint, save, and load the RL model
• Use RL research techniques in offline learning to train the model and evaluate
• Use Python decorators to deploy and serve the trained RL model using Ray Serve

Level: Beginners or intermediate data scientists/ML/RL engineers

Prerequisite knowledge or skills:

• Good skills in Python
• Basics in deep learning in TensorFlow2 or PyTorch
• Read this introduction to RL: https://www.anyscale.com/blog/an-informal-introduction-to-reinforcement-learning