Heads-Up No-Limit Texas Hold'em Poker (HUNL) Subgames

该数据集包含了四个开源的德州扑克单挑无限注子游戏，这些子游戏是在智能代理Libratus与顶级人类玩家对决期间实时解决的。这些子游戏具有不同的投注轮次和底池大小，其可利用性以每局游戏百万分之一大盲注（mbb/g）来衡量。数据集规模涵盖四个具有不同底池大小的子游戏，任务是用以在扑克子游戏的背景下评估CFR（反事实后悔最小化）算法。

This dataset comprises four open-source heads-up no-limit Texas Hold'em subgames, which were solved in real time during the head-to-head matches between the AI agent Libratus and top-tier human players. These subgames feature varied betting rounds and pot sizes, with their exploitability quantified in million big blinds per game (mbb/g). The dataset is designed to evaluate the CFR (Counterfactual Regret Minimization) algorithm within the context of poker subgames.