Reinforcement Learning agents that play the StarCraft II Minimaps and eventually will be able to play the full game.
This project depends on the sc2gym OpenAI Gym Environments.
This is still a work in progress. The train_a2c.py
no longer runs due to changes in the baselines
code. The current
plan is to replace both train_dqn.py
and train_a2c.py
with standalone implementations, and add a standalone A3C
implementation, as well as benchmark results on the various sc2gym
environments.