Linear Feature Encoding for Reinforcement Learning