Reinforcement Learning of Theorem Proving