Chris Pollett > Students >
Luo

    ( Print View)

    [Bio]

    [Blog]

    [CS 297 Proposal]

    [Deliverable 1]

    [Deliverable 2]

    [Deliverable 3]

    [Deliverable 4]

    [AI Methods for Dou Di Zhu-PDF]

    [Q-learning-PDF]

    [CS297 Report-PDF]

    [CS 298 Proposal]

    [CS298_Report-PDF]

    [Defense Slides-PDF]

CS298 Proposal

An AI for a Modification of Dou Di Zhu

Xuesong Luo (xuesong.luo@sjsu.edu)

Advisor: Dr. Chris Pollett

Abstract:

Dou Di Zhu is a famous card game in China. This card game uses 54 cards (include 2 Jokers), and it needs 3 players for 2 sides. One side is called the Landlord and consist of one player, the other side is called the peasants and consist of 2 players. Each player receives 17 random cards, and there are 3 extra random cards that belong to the Landlord. There are some research papers already used the Decision Tree and the Rule Base on the Dou Di Zhu game. My project aims to create a multiple online card game: Dou Di Zhu, and train the Q-learning with Neural Network AI that can join the game to play with humans.



CS297 Results:

1. An online multiplayer game: Dou Di Zhu

2. Q-learning and card game

3. Develop a Q-learning algorithm based on the Dou Di Zhu

4. Re-implement an existing AI technique of Dou Di Zhu



Schedule:

Week 1: Jan 23 - Jan 28First meeting, figuring out what to do to enhance the project.
Week 2-7: Jan 29 - Mar 10Learning how to combine the Q-learning with Neural Network and implement my AI model of Dou Di Zhu with Neural Network.
Week 8-11: Mar 11 - Apr 07Modify the AI Dou Di Zhu game that can play with the human players and the game UI.
Week 12-16: Apr 08 - May 05Prepare the project report and slides for review


Innovations and Challenges:

  • Figure out how to combine the Q-learning algorithm with Neural Network.
  • Modify the Q-learning algorithm with Neural Network model have a higher winning rate than average in the Dou Di Zhu card game.
  • Ensuring the last AI model can play with the human players.


Key Deliverables:

Software
  • Design and create a basic Q-learning algorithm combined with Neural Network.
  • Training a succeed Q-learning with Neural Network AI for the Dou Di Zhu which can play with humans.
  • Re-design UI of the Dou Di Zhu game beyond the old one.
Report
  • CS 298 report.
  • CS 298 presentation.


References:

[2019] Yang You, Liangwei Li, Baisong Guo, Weiming Wang, Cewu Lu. "Combinational Q-Learning for Dou Di Zhu", arXiv:1901.08925

[2017] Renzhi Wu, Shuai Liu, Shuqin Li, Meng Ding. "The design and Implementation of a Computer Game Algorithm of Do Dizhu", IEEE

[Unknown] Zhennan Yan, Xiang Yu, Tinglin Liu, Xiaoye Han. Fight the Landlord (Dou Di Zhu).