/pysc2-a3c

A StarCraft II / pysc2 agent implement by a3c(Asynchronous Advantage Actor-Critic)

Primary LanguagePython

Introduction

It's a research about reinforcement learning. I use pysc2 as environment.The original paper is sc2le, you can see and know how it work, and algorithm is a3c.

Requirement

  1. of course. the StarCraft II game, and remember to download the mini game
  2. pysc2 (version is 2.0.1)
  3. numpy
  4. tensorflow-gpu (version is 1.10)
  5. absl-py

How to run

Because my system is windows 10 and anaconda of python3.7, so I create a python environment named "tf-gpu" and use python3.5 to run, you can skip the first and the last step if you are using another system.

  1. activate tf-gpu
  2. cd my-pysc2
  3. python main.py
  4. deactivate

testing

python main.py --map=MoveToBeacon --train=False --envs=1 --exploration_mode=eg

Reference

The code is based on here. I really appreciat.

Apologize

I'm not familiar with tensorflow and numpy, and just learning reinforcement learning. So the code has many annotation.