Convergent Combinations of Reinforcement Learning with Linear Function Approximation