Reinforcement Learning From Scratch