Deep Q-learning in Keras

Now that we understand how to train an agent to select optimal state action pairs, let's try to solve a more complex environment than the taxi cab simulation we dealt with previously. Why not implement a learning agent to solve a problem that was originally crafted for humans themselves? Well, thanks to the wonders of the open source movement, that is exactly what we will do. Next on our task list, we will implement the methodologies of Mnih et al. (2013, and 2015) referring to the original DeepMind paper that implemented a Q-learning based agent. The researchers used the same methodology and neural architecture to play seven different Atari games. Notably, the researchers achieved remarkable results for six of the seven different games it was tested on. In three out of these six games, the agent was noted to outperform a human expert. This is why...