Reinforcement Learning Basis Soap