https://towardsdatascience.com/reinforcement-learning-generalisation-on-continuing-tasks-ffb9a89d57d0