Skip to content

Latest commit

 

History

History
43 lines (34 loc) · 2.57 KB

README.md

File metadata and controls

43 lines (34 loc) · 2.57 KB
Raw

AIWolfPy

Create python agents that can play Werewolf, following the specifications of the AIWolf Project

This has been forked from the official repository by the AIWolf project, and was originally created by Kei Harada.

Changelog:

Version 0.4.9a

  • Added support material in English

Version 0.4.9

  • Changed differential structure (diff_data) into a DataFrame

Version 0.4.4

  • removed daily_finish
  • Added update callback (with request parameter)
  • Connecting is now done through a instance, not a class

Version 0.4.0

  • Support for python3
  • Made file structure much simpler

Running the agent and the server locally:

  • Download the AIWolf platform from the [AIWolf public website] (http://www.aiwolf.org/server/)
    • Don't forget that the local AIWolf server requires JDK 11
  • Start the server with ./StartServer.sh
    • This runs a Java application. Select the number of players, the connection port, and press "Connect".
  • In another terminal, run the client management application ./StartGUIClient.sh
    • Another Java application is started. Select the client jar file (sampleclient.jar), the sample client pass, and the port configured for the server.
    • Press "Connect" for each instance of the sample agent you wish to connect.
  • Run the python agent from this repository, with the command: ./python_sample.py -h [hostname] -p [port]
  • On the server application, press "Start Game".
    • The server application will print the log to the terminal, and also to the application window. Also, a log file will be saved on "./log".
  • You can see a fun visualization using the "log viewer" program.

Running the agent on the AIWolf competition server:

  • After you create your account in the competition server, make sure your client's name is the same as your account's name.
  • The python packages available at the competition server are listed in this page
  • You can expect that the usual packages + numpy, scipy, pandas, scikit-learn are available.
    • Make sure to check early with the competition runners, specially if you want to use something like an specific version of tensorflow.
    • The competition rules forbid running multiple threads. Numpy and Chainer are correctly set-up server side, but for tensorflow you must make sure that your program follows this rule. Please see the following post
  • For more information, a tutorial from the original author of this package can be seen in this slideshare (in Japanese).