In Monte Carlo exploration starts, we explore all state-action pairs and choose the one that gives us the maximum value. But think of a situation where we have a large number of states and actions. In that case, if we use the MC-ES algorithm, then it will take a lot of time to explore all combinations of states and actions and to choose the best one. How do we get over this? There are two different control algorithms. On policy and off policy. In on-policy Monte Carlo control, we use the ε greedy policy. Let's understand what a greedy algorithm is.
A greedy algorithm picks up the best choice available at that moment, although that choice might not be optimal when you consider the overall problem. Consider you ...
