4. Getting Started with OpenAI and TensorFlow for Reinforcement Learning

Overview

This chapter introduces you to some key technologies and concepts to get started with reinforcement learning. You will become familiar with and use two OpenAI tools: Gym and Universe. You will learn how to deal with the interfaces of these environments and how to create a custom environment for a specific problem. You will build a policy network with TensorFlow, feed it with environment states to retrieve corresponding actions, and save the policy network weights. You will also learn how to use another OpenAI resource, Baselines, and use it to train a reinforcement learning agent to solve a classic control problem. By the end of this chapter, you will be able to use all the elements we will introduce to build and train an agent to play a classic Atari video game, thus achieving better-than-human performance.