Reinforcement Learning on Atari

A WIP implementation of Rainbow (https://arxiv.org/abs/1710.02298).

After installing the requirements, you also need to run pip install gym[atari] manually. If you are using CUDA also make sure to install the right wheel from https://pytorch.org/.