In the previous chapters, we have learned about fundamental concepts of reinforcement learning (RL) and several RL algorithms, as well as how RL problems can be modeled as the Markov Decision Process (MDP). We have also seen different model-based and model-free algorithms that are used to solve the MDP. In this chapter, we will see one of the classical problems in RL called the multi-armed bandit (MAB) problem. We will see what the MAB problem is and how to solve the problem with different algorithms followed by how to identify the correct advertisement banner that will receive most of the clicks using MAB. We will also learn about contextual bandit that is widely used for building recommendation systems.

In the chapter, you will learn about the following:

• The MAB problem
• The epsilon-greedy algorithm
• The softmax exploration algorithm
• The upper confidence bound algorithm
• The Thompson sampling algorithm
• Applications of MAB
• Identifying the right advertisement banner using MAB
• Contextual bandits