Portal de Periódicos da CAPES

Odds Estimating with Opponent Hand Belief for Texas Hold’em Poker Agents

2021; Springer Science+Business Media; Linguagem: Inglês

10.1007/978-3-030-93046-2_5

1611-3349

Zhenzhen Hu, Jing Chen, Wanpeng Zhang, Shaofei Chen, Weilin Yuan, Junren Luo, Jiahui Xu, Xiang Ji,

Gambling Behavior and Treatments

Along with the great success of superhuman AI in succession, the development of Texas Hold’em poker agents is entering a new stage. The efforts to create indefectible AI transfers to develop new AIs which can exploit opponents and explain its own decisions better. Hand odds estimating used to state abstracting, situation evaluating, decision assisting is one of the key foundations for such new agents. But all the current methods implicitly assume a uniform distribution over the cards the opponent could be holding, which makes the win rate over-evaluated. In this paper three hand odds estimating methods considering the opponent hand belief are proposed to cope with this problem. We suggest the expected win rate algorithms with start hand range (EWR-SHR) and expected win rate algorithm with fold rate (EWR-FR) for preflop round and flop/turn/river round respectively. These two algorithms predict the opponent’s hand range based on the opponent model and observed action not fold, their additional computation complexity is O(1). The expected win rate algorithm with opponent hand distribution (EWR-HD) is the third method suitable for all rounds which uses the opponent model and observed action check/call/raise to infer the distribution of the opponent hand cards. Their features are compared, usages are summarized, and the experiment result indicates that all of them can evaluate the game situation more precisely than the current methods.