Approximations of the Restless Bandit Problem