Skip to content

根据一些人工智能的一些经典算法,以及目前人工智能在星际争霸II的表现,通过官方API,我们制作了相应的mini-game。同时,我们也发明了与Q-value延伸出来的新算法,大大降低复杂度,去优化机器在人机对战中的表现。

Notifications You must be signed in to change notification settings

fanbbbb/Large-Action-Space-Reinforcement-Learning-StarCraft-II

Repository files navigation

AI-StarCraft-II

Final project of CS181 in ShanghaiTech. Focus on experiments with different opponent setting policy with basis of DQN method on StarCraft 2.

Teammate: Keyi-Yuan, Qin-QI, Ruiqi-Liu, Yintao-Xu

For the reason that we do slight modification to the pysc2, i include pysc2 lib in this repository(pysc2) under Apache License 2.0.

Quick Start

  1. git clone https://github.com/liubai01/AI-StarCraft-II.git
  2. Copy map/SimpleContest.SC2Map to your map directory. (e.g: on my PC, it is: D:\billizard\StarCraft II\Maps\Melee, it depends on where you install the game)
  3. IMPORTANT: Modify the log file path in lib\pysc2_info_saver.py, the recorder class! (留言:帮我跑程序的话一定要设置这个,否则实验白做了。。)

Documents

TBD, for now, there are only some scratches

  1. action space

  2. feature engineering-observation

Other scratches: https://github.com/Q71998/yascai

Reference documentation

General idea of pysc2: [Portal]

Environment in pysc2: [Portal]

Update Log

2018/11/15: backbone of this project. Realize detection the center of build(currently command center) by lib/building/get_building_center(<obs>, <uid>). Realize basic agent in agent_Alpha.

2018/12/19: backbone of the self-game RL learning setting

About

根据一些人工智能的一些经典算法,以及目前人工智能在星际争霸II的表现,通过官方API,我们制作了相应的mini-game。同时,我们也发明了与Q-value延伸出来的新算法,大大降低复杂度,去优化机器在人机对战中的表现。

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages