Policy iteration solves a finite Markov Decision Process by alternating two steps until the greedy policy stops changing.
