Towards Finding a Solution for Collectible Card Games with Counterfactual Regret Minimization

• Haoyu, Zhang

• Yoshimasa Tsuruoka

Abstract

In recent years, agents trained to cope with various environments have received attention in the machine learning field. An intuitive rational decision-making environment is card games and researchers are especially focused on solving the game of poker. In this work, we focus on Collectible Card Games (CCGs) which have rich diversity of game play and much more randomness compared to poker. In order to solve such complicated card games, we create several simple models of the real game and conduct experiments on these models with Counterfactual Regret Minimization (CFR) algorithms. As results, agents learn equilib-rium strategies successfully in the simple model and we confirm the convergence of Counterfactual Regret Minimization (CFR) algorithms.

In recent years, agents trained to cope with various environments have received attention in the machine learning field. An intuitive rational decision-making environment is card games and researchers are especially focused on solving the game of poker. In this work, we focus on Collectible Card Games (CCGs) which have rich diversity of game play and much more randomness compared to poker. In order to solve such complicated card games, we create several simple models of the real game and conduct experiments on these models with Counterfactual Regret Minimization (CFR) algorithms. As results, agents learn equilib-rium strategies successfully in the simple model and we confirm the convergence of Counterfactual Regret Minimization (CFR) algorithms.

Journal

• ゲームプログラミングワークショップ2019論文集

  ゲームプログラミングワークショップ2019論文集 2019 235-241, 2019-11-01

Details

• CRID: 1050855522065599360
• NII Article ID: 170000180601
• Web Site: http://id.nii.ac.jp/1001/00199908/
• Text Lang: en
• Article Type: conference paper
• Data Source:
  • IRDB
  • CiNii Articles