Thus far, we've learned how we can use reinforcement learning methods to play board games utilizing UCT and MCTS; now, let's see what we can do with video games. In Chapter 8, Reinforcement Learning, we saw how we could use reinforcement learning methods to complete basic tasks such as the OpenAI cartpole challenge. In this section, we'll be focusing on a more difficult set of games: classic Atari video games, which have become standard benchmarks for deep learning tasks.
You might be thinking – can't we extend the methods that we used in the cartpole environment to Atari games? While we can, there's a lot more input that we have to handle. In Atari environments, and really any video game environment, the inputs to the network are individual pixels. Instead of the simple four control variables for cartpole, we are now dealing...