Speeding up the learning of equivalence classes of Bayesian network structures

For some time, learning Bayesian networks has been both feasible and useful in many problems domains. Recently research has been done on learning equivalence classes of Bayesian networks, i.e. structures that capture all of the graphical information of a group of Bayesian networks, in order to increase learning speed and quality. However learning speed still remains quite slow, especially on problems with many variables. This work aims to describe a method to speed up algorithm learning speed. A brief overview of learning Bayesian networks is given. A method is then given, so that tests of whether a particular move is valid can be cached. Finally, experiments are conducted, which show that applying this caching method produces a marked increase in learning speed.