Here we simulate k-arm bandit problem, which is one of the basic reinforcement learning problems. It's a simple excercise which illustrates the exploration-exploitation tradeoff. We experimented training our agent with greedy (which always maximizes immediate reward) and epsilon-greedy (which mostly maximizes immediate reward but also occasionaly takes risk for exploring).

This particular excercise was taken from a textbook by Sutton and Barto.