For this chapter, we continued exploring TD learning. We looked at an example of an online TD (0) method called SARSA. Then, we looked at how we can discretize an observation space to tackle harder problems but still use the same toolset. From there, we looked at how we could tackle harder continuous space problems such as CartPole. After that, we revisited TDL and then looked to n step forward views, decided that was less than optimal, and then moved to backward views and eligibility traces, which led to us uncovering TD (λ), SARSA(λ), and Q (λ). Using SARSA(λ), we were able to solve the MountainCar environment in far less time. Finally, we wanted to tackle a far more difficult environment, LunarLander using SARSA(λ) without deep learning.
In the next chapter, we look at introducing deep learning and escalate ourselves to deep reinforcement...
