The Information Model for a Class of Imperfect Information Game

With the increasing research works on imperfect information games(IIGs), different kinds of Imperfect information games have been studied which vary a lot in their information properties. Generally, the information sets and the relations between them decide an IIGs information property, but sometimes the game tree can be transformed to a more efficient form—the “flatten” form. The flatten form needs less memory and can accelerate the search process. The flatten form can describe the relations between information sets but it has difficulty in recording the composition of information sets, so how to represent and manage the composition information of information sets in the flatten form turns to be a new problem. In this paper, a new concept of IIG—imperfect information space is introduced and two types of IIGs are studied. Then, a novel general information model based on bipartite graph is proposed. With the help of information model, we study the information acquisition problem and use the Markov network to manage information. The Markov network learns the dependency of the attitudes in our information model from the archives of human games automatically, that helps us to build the information model without expert knowledge and makes the model to be more impersonal. The experiments on Siguo game show the effectiveness of our general information model on the acquisition and management of imperfect information, and also prove the efficiency of the Markov network.