An MRP Formulation for Supervised Learning: Generalized Temporal Difference Learning Models