ABSTRACT
Bayesian network structure learning concerns learning the DAG in a Bayesian network from data. The Bayesian score is the probability of the data D given the DAG. Recall that Bayesian networks represent a large joint distribution of random variables succinctly. The chapter presents an algorithm in which the search space is the set of all DAGs containing n nodes, where n is our number of random variables. It also presents an algorithm that searches over DAG patterns. The chapter discusses a quite different structure learning technique called constraint-based learning. It illustrates the constraint-based approach by showing how to learn a DAG faithful to a probability distribution. The machine learning community developed a quite different method, called class probability trees, for handling discrete target variables. The chapter presents applications specifically concerned with learning Bayesian networks from data.
