After two years of getting more into this field of RL research, I realized this reproduction was not even close to the original implementation. I would recommend https://github.com/simonmeister/pysc2-rl-agents as a better reproduction. I am archiving this repo.

You may view the Jupyter Writeup here. In this project, we

• Utilized Openai Gym to interact with Pysc2, the SC II Learning Environment
• Successfully incorporated Tensorforce, a reinforcement learning framework, to train SC II game agents
• Constructed maintainable and understandable machine learning code