reinforcement learning by openai gym, pacman sample (1750k steps)
Explanation : https://qiita.com/kznx/items/a59fa5b73c32cbda33d9