Statistics of Computing near-optimal policies from trajectories by solving a sequence of standard supervised learning problems