A New Method for tackling Asymmetric Decision Problems
Chain Event Graphs are probabilistic graphical models designed especially for the analysis of discrete statistical problems which do not admit a natural product space structure. We show here how they can be used for decision analysis, and describe an optimal decision strategy based on an efficient local computation message passing scheme. We briefly describe a method for producing a parsimonious decision CEG, analogous to the parsimonious ID, and touch upon the CEG-analogues of Shachter's barren node deletion and arc reversal for ID-based solution.