Anticipatory Learning for Focusing Search in Reinforcement Learning Agents