PyTorch 1.x Reinforcement Learning Cookbook

Solving Multi-armed Bandit Problems

Multi-armed bandit algorithms are probably among the most popular algorithms in reinforcement learning. This chapter will start by creating a multi-armed bandit and experimenting with random policies. We will focus on how to solve the multi-armed bandit problem using four strategies, including epsilon-greedy, softmax exploration, upper confidence bound, and Thompson sampling. We will see how they deal with the exploration-exploitation dilemma in their own unique ways. We will also work on a billion-dollar problem, online advertising, and demonstrate how to solve it using a multi-armed bandit algorithm. Finally, we will solve the contextual advertising problem using contextual bandits to make more informed ...