Reinforcement Learning for Symbolic Mathematics