Reinforcement Learning for Atari Games