A MCTS AI we created for an university course. It plays the famous game, Kalah.

This project is developed by Feifan Chen(陈非凡), Xiaotan Zhu(朱笑谈), Yirong Yu(俞奕戎), and Wenqing Zong(宗文卿).

See a brief introduction here. But in this project, we are playing the 7 * 7 version. Pie rule is applied here to cancle out first player advantage.

We used MCTS algorithm with some improvement to implement our AI agent.

Our improvement follows these research papers:

• An analysis for strength improvement of an MCTS-based program playing Chinese dark chess

• Case Study - Kalah

• Parallel Monte-Carlo Tree Search

• MCTS-Minimax_Hybrids

• Trade-Offs in Sampling-Based Adversarial Planning

• Monte Carlo Tree Search with heuristic evaluations using implicit minimax backups - IEEE Conference Publication

1. Please download and open our project in IntelliJ and accept all default settings.

2. Open IntelliJ, click the green "Run" button.

3. Terminate it, we merelly want IntelliJ to build our code.

4. In Test_Agents folder, run ./compile.sh for MacOS and Linux, or .\compile.bat for Windows.

5. • If you want to see two agents playing agaist each other, in Test_Agents folder, do:

   java -jar ManKalah.jar "java -jar <FirstAgent>" "java -jar <SecondAgent>"

   where <FirstAgent> and <SecondAgent> can be (and must be) replaced by a jar file in Test_Agents folder.

   • If you are lazy and you are using MacOS or Linux, simply do:

     ./run

     This will test our AI agent again all four test agents with both being player 1 and player 2.

   • If you want to play against our agent (or any other agent), open a ternimal, do:

     nc localhost 12345

     and then in Test_Agents folder, do:

     java -jar ManKalah.jar "java -jar <Agent>" "nc localhost 12345"

     The first argument indicates the first player, the second indicates the second player. For our protocal, please see doc folder for more details.