Deepstack for Bucket

This repository contains the code for running the experiments for the following paper:

Ganzfried S , Sandholm T . Potential-aware imperfect-recall abstraction with earth mover's distance in imperfect-information games[C]// Twenty-eighth Aaai Conference on Artificial Intelligence. AAAI Press, 2014.

This code is designed to:

• Implement an abstract method to significantly improve performance on no-limit Texas Hold'em.
• Apply and test the effect of this abstract method on the Deepstack algorithm.

Dependencies

• Python 3
• Pyemd
• Numpy
• Scipy
• Scikit-learn
• Matplotlib

How to Run

• generate data

  python Generate_data.py [OPTIONS]

  • Options

  Name, shorthand	Default	Range	Description
  --street	river	river | turn | flop	the round name
  --file_path	data/	-	the relative path for storing data

• Clustering data

  ·python Cluster_data.py [OPTIONS]

  • Options

  Name, shorthand	Default	Range	Description
  --street	river	river | turn | flop	the round name
  --file_path	data/	-	the relative path for storing data
  --k	5	Positive integer	the number of clusters
  --initialMethod	kmean++	kmeans++ | random	initialize cluster center point method
  --ifsave	True	True | False	whether to save the cluster center point

• Save the correspondence between the hand and the bucket

  python Cluster_result.py

• Visual clustering results

  python Data_Visualization.py [OPTIONS]

  • Options

  Name, shorthand	Default	Range	Description
  --street	river	river | turn | flop	the round name
  --mode	test	data | results | test	plot data mode
  --ifsave	True	True | False	whether to save the cluster center point

  • Details

    data mode : Plot raw data distribution

    results mode : Plot bucket data distribution

    test mode :Plot both raw and bucket data distribution

• Check the minimum cluster center point EMD distance

  python Check_MinEMD.py

Algorithm

This is the first algorithm for computing potential-aware imperfect-recall abstractions, using EMD as the distance metric. Experiments on no-limit Texas Hold’em show that our algorithm leads to a statistically significant improvement in performance over the previously best abstraction algorithm.

If you want to know more details, please read this paper