Research On Machine Game Model And Algorithms Of Bridge Bidding

Studying card machine game can promote decision-making and control in the context of incomplete information in economy and society.As one of the most complex card games,bridge is divided into bidding and playing.This paper studies machine game algorithm for bidding.In the traditional bidding machine game research,Monte Carlo sampling based on expert experience,as the best technology at present,has faced development barriers and is difficult to achieve breakthroughs.This paper creatively decomposes the bidding problem domain into three sub-problem domains with evolutionary relationship:first bidding,no-contention bidding and contention bidding,and uses deep learing and reinforcement learning to study them respectively,so as to reduce the research gradient,making the machine understand the meaning of bidding and get rid of the restriction of human bidding experience.This paper takes the massive data generated by online platform players of Xinrui Bridge Company under the standard natural system CCBA of China Bridge Association as the research fulcrum,and makes the model achieve first bidding under CCBA system as the basic goal,and then evolves the model from the first bidding problem domain to the non-contention bidding and contention bidding problem domain,tries to optimize the model action strategy with reinforcement learning method.Aiming at the key problems faced by bidding machine game research,such as incomplete information characteristics,learning bidding system correctly,information representation of hand cards and bidding sequence,accurate expression of bidding transmission information,general orientation,information discretization,particularity of PASS action,cooperation and game and so on,this paper designs seven different input layers of neural network,each of which solves several key problems.Then,based on the input layer of seven kinds of neural networks,the first bidding algorithm,no-contention bidding algorithm and contention bidding algorithm are designed and implemented.The results show that the first bidding algorithm achieves the design purpose,no-contention bidding algorithm and contention bidding algorithm have poor performance.The first bidding algorithm and no-contention bidding algorithm have the best performance in the functional one-dimensional placeholder input layer,and the contention bidding algorithm has the best performance in the three-dimensional placeholder input layer.Finally,according to the results of the algorithm implementation,the possible problems are analyzed,and the future direction of work is determined according to the problems.