ABSTRACT
In the previous chapter we addressed the problem of parameter learning in a Bayesian network. In this chapter we discuss structure learning. Section 13.1 introduces the problem of Bayesian network structure learning from data. Sections 13.2 and 13.3 discuss two different techniques for learning that structure. Then Section 13.4 provides a large-scale example of structure learning. Next,
Table 13.1 Data on 12 Workers
Case Sex Height (inches) Wage ($) 1 female 64 30,000 2 male 64 30,000 3 female 64 40,000 4 female 64 40,000 5 female 68 30,000 6 female 68 40,000 7 male 64 40,000 8 male 64 50,000 9 male 68 40,000 10 male 68 50,000 11 male 70 40,000 12 male 70 50,000
Section 13.5 introduces software packages for learning structure, and Section 13.6 shows how we can use Bayesian network structure learning to learn something about causal influences. Finally, Section 13.7 concerns class probability trees, which are used for learning structure when we have a single target variable.
