Using Temporal Neighborhoods to Adapt Function Approximators in Reinforcement Learning