In this chapter, we first introduced ourselves to the world of RL. We looked at what makes RL so unique and why it makes sense for games. After that, we explored the basic terminology and history of modern RL. From there, we looked to the foundations of RL and the Markov decision process, where we discovered what makes an RL problem. Then we looked to building our first learner a value learner that calculated the values of states on an action. This led us to uncover the need for exploration and exploitation and the dilemma that constantly challenges RL implementers. Next, we jumped in and discovered the full Q-learning equation and how to build a Q-learner, where we later realized that the full Q equation was beyond what we needed for our unconnected state environment. We then reverted our Q learned back into a value learner and watched it solve the contextual bandit problem.
In the next chapter, we will continue from where we left off and look into how rewards are discounted with the Bellman equation, as well as look at the many other improvements dynamic programming has introduced to RL.
