A general approach to learn a network structure is to heuristically search the space of network structures for the one that best fits a given data set. The key to the search is a score function which evaluatos different network structures. In this paper,we give a detailed introduction to two representative score functions: Bde,MDL. We also discuss two widely used learning algorithms that apply those score functions to direct their search:hill climbing and simulated annealing. Finally, We briefly sketch an algorithm,SEM ,that can learn a network structure from incomplete data.
展开▼
