RL-SC2 is a DRL learning environment for DeepMind's PySC2 interface. It's specifically tailored down for reinforcment learning, with essential components like replay buffer and trajectory generating by interacting with PySC2. It can be used as a boilerplate to speed up your own design of agents.
The ultimate goal is to reproduce DeepMind's paper, which shows super-human performance in real-time playing. Since there is no official code published, mistakes and differences are expected in this project.