Discovering Logical Formulaic Alphas using Deep Reinforcement Learning #