Introduction

I am trying to implement AlphaStar based on supplementary material of DeepMind.

Reference

1. Download replay file(4.8.2 version file is needed): https://github.com/Blizzard/s2client-proto/tree/master/samples/replay-api
2. Extracting observation, action from replay file: https://github.com/narhen/pysc2-replay
3. Transfomer network of Tensorflow 2.0: https://www.tensorflow.org/tutorials/text/transformer
4. Resblock of Tensorflow 2.0: https://zhuanlan.zhihu.com/p/87047648

Version

1. Python3
2. PySC2 3.0.0
3. Tensorflow 2.2.0

Running test

To check you setting all your computer environment correctly. Run env_test.py file in your terminal.

Then, screen of PySC2 will start and you can see some activation of rule-based agent of Terran.

Detailed information

I am writing explanation for code at Medium as series.

1. Tutorial about replay file: https://medium.com/@dohyeongkim/alphastar-implementation-serie-part1-606572ddba99
2. Tutorial about agent class: https://medium.com/@dohyeongkim/alphastar-implementation-serie-part2-3edced5df00b
3. Tutorial about encoder network: https://medium.com/@dohyeongkim/alphastar-implementation-series-part3-d315d2ad5a3
4. Tutorial about head network: https://dohyeongkim.medium.com/alphastar-implementation-series-part4-ee64bb93fe59
5. Tutorial about training network: https://dohyeongkim.medium.com/alphastar-implementation-series-part5-fd275bea68b5

Preprocess observation of PySC2

I am adding the necessary code for running AlphaStar based on psuedocode of DeepMind paper

1. Extract information from Replay file: trajectory.py file
2. Encoder, Core, Head network: network.py file
3. Preprocessing function : utils.py file
4. Edited unit inforamtion of PySC2: units_new.py file
5. Edited upgrade inforamtion of PySC2: upgrades_new.py file

You can run AlphaStar application by running alphastar.py file after putting all additional file in same folder.

$ python alphastar.py