Foundations of Deep Reinforcement Learning: Theory and Practice in Python

13. Hardware

Deep RL owes part of its success to the arrival of powerful hardware. To implement or use deep RL algorithms, one inevitably needs to understand some basic details about a computer. These algorithms also require a significant amount of data, memory, and computational resources. When training agents it is useful to be able to estimate an algorithm’s memory and computing requirements and to manage data efficiently.

This chapter is intended to build intuition for the types of data encountered in deep RL, their sizes, and how to optimize them. We first briefly describe how hardware components such as CPU, RAM, and GPU work and interact. Then, in Section 13.2 we give an overview of data types. Section 13.3 discusses the common types ...