Here is a deep Q-network implemented in Pytorch that has learned to play classic Atari games, breakout and pong, with the python Gym library. Both agents trained in ~1 hour on an NVIDIA 1070ti.
The atari agent is able to consitently clear all but a few blocks, while it sometimes is able to clear all. The pong agent (green) can consitently win against the built-in Atari AI.
