OpenHoldem
收藏资源简介:
OpenHoldem是由中国科学院自动化研究所开发的一个用于大规模不完美信息游戏研究的基准工具,主要应用于无限制德州扑克(NLTH)。该数据集提供了三种主要组件:评估协议、强大的基准NLTH AI和在线测试平台。评估协议包括多种评估指标,用于全面测试不同NLTH AI的性能。基准AI包括基于规则的AI、基于CFR的静态AI、类似DeepStack的在线AI和基于深度强化学习的AI。在线测试平台允许研究人员通过易于使用的API将其AI与内置基准进行比较,从而促进对未解决的理论和计算问题的进一步研究,如对手建模和人机交互学习。
OpenHoldem is a benchmark tool developed by the Institute of Automation, Chinese Academy of Sciences for research on large-scale imperfect-information games, primarily applied to No-Limit Texas Hold'em (NLTH). This dataset offers three core components: an evaluation protocol, a powerful suite of benchmark NLTH AI systems, and an online testing platform. The evaluation protocol encompasses multiple evaluation metrics for comprehensively assessing the performance of diverse NLTH AI systems. The benchmark AI suite includes rule-based AI, static CFR-based AI, online DeepStack-like AI, and deep reinforcement learning-based AI. The online testing platform enables researchers to compare their own AI systems with the built-in benchmarks via a user-friendly API, thereby facilitating further research on unsolved theoretical and computational problems such as opponent modeling and human-computer interactive learning.



