Reinforcement Learning in a Nonstationary Environment: The El Farol Problem