Title: An integration model for Texas Hold'em

Authors: Yajie Wang; Shengyu Han; Zhihao Wei; Zhonghui Shi

Addresses: Engineering Training Centre, Shenyang Aerospace University, Shenyang City, Liaoning Province, 110000, China ' Department of Computer Science, Shenyang Aerospace University, Shenyang City, Liaoning Province, 110000, China ' Department of Computer Science, Shenyang Aerospace University, Shenyang City, Liaoning Province, 110000, China ' Department of Computer Science, Shenyang Aerospace University, Shenyang City, Liaoning Province, 110000, China

Abstract: Texas Hold'em is a representative of an incomplete information game. Existing research on computing Nash equilibrium as a Texas Hold'em strategy has problems, including high resource consumption and conservative strategies. To solve the above problems, an integration model combining deep learning and reinforcement learning is proposed. Firstly, to reduce the storage resources consumed due to the large Texas Hold'em state space, a long short-term memory (LSTM) is designed to predict the game results. Since the win rate and historical action information are used as input data by the LSTM, a convolutional neural network (CNN) is designed to predict the current win rate. Secondly, in order to enable the strategy to have dynamic adjustment ability, the deep Q-network(DQN) is used to generate the strategy by using the results predicted by LSTM. Finally, an agent is implemented to provide training data for LSTM. The experimental results show that the model wins more chips, which proves that it can be used as a solution for incomplete information games.

Keywords: Texas Hold'em; reinforcement learning; deep learning; DQN algorithm; integration model.

DOI: 10.1504/IJCSM.2023.134575

International Journal of Computing Science and Mathematics, 2023 Vol.18 No.3, pp.203 - 213

Received: 18 Jul 2022
Accepted: 08 May 2023
Published online: 27 Oct 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article