A Neighborhood Correlated Empirical Weighted Algorithm for Fictitious Play

Abstract

Fictitious play is a widely used learning model in games. In the fictitious play, players compute their best replies to opponents’ decisions. The empirical weighted fictitious play is an improved algorithm of the traditional fictitious play. This paper describes two disadvantages of the empirical weighted fictitious play. The first disadvantage is that distribution of the player’s own strategies may be important to make a strategy as times goes. The second is that all pairs of players selected from all players ignore their neighborhood information during playing games. This paper proposes a novel neighborhood correlated empirical weighted algorithm which adopts players’ own strategies and their neighborhood information. The comparison experiment results demonstrate that the neighborhood correlated empirical weighted algorithm can achieve a better convergence value.