Skip to content
This repository has been archived by the owner on Dec 25, 2021. It is now read-only.

Latest commit

 

History

History
44 lines (36 loc) · 1.9 KB

README.md

File metadata and controls

44 lines (36 loc) · 1.9 KB
Raw

CGI Go Intelligence (abbr. CGI)

This repository introduces the usage of running the computer Go program CGI (for version 4.0). CGI 4.0 is a version based on the AlphaZero algorithm and improved the training process by the population based training. For the background history of other CGI versions, please refer to https://cgilab.nctu.edu.tw/~icwu/aigames/CGI.html.

Citing CGI

If you use CGI for research, please consider citing our paper as follows:

@inproceedings{wu2020accelerating,
  title={Accelerating and improving alphazero using population based training},
  author={Wu, Ti-Rong and Wei, Ting-Han and Wu, I-Chen},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={34},
  number={01},
  pages={1046--1053},
  year={2020}
}

Usage

Environment

We recommend running CGI via the provided container. Here we provide two ways to set up the environment as follows:

  1. You can directly download the built image from the Docker Hub with the following commands:
docker pull kds285/cgigo
  1. You can also build the container via the Dockerfile yourself:
docker build cgigo .

Running CGI

We provide an example configuration, named cgi_example.cfg, to run CGI. The example configuration will use 1 GPU and run 800 simulation counts for each move. Following are the ways to build and run CGI:

  1. Build the program via ./scripts/setup-cmake.sh release caffe2 && make
  2. Run the program via Release/CGI -conf_file cgi_example.cfg

After the program is started, you can interact with the program via the GTP protocol.

You can also edit the configuration file for different settings as follows:

  • let the program play by simulation count: SIM_CONTROL=COUNT
    • adjust the simulation count: SIM_COUNT_LIMIT=800
  • let the program play by thinking time: SIM_CONTROL=TIME
    • adjust the thinking time: SIM_TIME_LIMIT=1, note that the number indicates seconds