Local Structural Learning of Causal Networks

• 作 者: GENG Zhi WANG Chang-chang ZHOU You
• 机 构: CenterforStatisticalScience,PekingUniversity
• 出 处: 2010年第十届中日统计研讨会
• 发表日期: 2010年10期

　　In some applications,we may be interested in discovering a local causal network around a target variable rather than the whole network over all variables.For example,we want to predict the target in the cases with external interventions,or we are interested in direct and indirect causes of a disease and further discriminate direct causes from other indirect causes.There are many algorithms for structural learning,but most of them are for constructing a whole network over all variables,such as Pearl (2000),Spirtes et al.(2000),Heckerman (1999),Tsamardinos et al.(2006),Xie et al.(2006) and Xie and Geng (2008).To discover a local causal network,however,it is inefficient to construct the whole network over a large number of variables.Correlations between variables are useful for prediction in the case that individuals to be predicted come from the same population as the training data.