LeadGenius: Mastering opening leads in the game of bridge

Bridge is an imperfect-information card game where the opening lead plays a crucial role in determining the game's outcome. Existing algorithms for Bridge opening leads often rely on Monte Carlo search with random sampling, which do not fully utilize information from the bidding phase and thus might result in unsatisfied opening-lead decisions. This paper introduces a novel Bridge opening-lead algorithm called LeadGenius, which leverages information from the bidding phase to guide sampling strategies, thereby enhancing decision-making ability in the opening lead. Specifically, LeadGenius first generates hand probability distributions through neural networks with supervised learning. In the meantime, critical hand constraints of players are identified by analyzing bidding sequences. Then, based on derived hand probability distributions and hand constraints, an algorithm of priority-based sampling with filtering is developed for hand sampling. Finally, an opening lead is made based on sampled hands via a search algorithm named Perfect Information Monte Carlo. Experiments demonstrate that LeadGenius outperforms championship-winning Bridge software Wbridge5 and tournament-level experts.