abn: Data Modelling with Additive Bayesian Networks

This library provides computational routines to help determine optimal Bayesian Network models for a given data set, where these models are used to identify all statistical dependencies present in messy, complex data. The usual term used to describe this model selection process is structure discovery, or more generally, data mining. Currently, a standard heuristic search and order based exact search are implemented, across two different types of Bayesian Network model: i) the classical (conjugate) contingency formulation for observations comprising of binary or multinomial variables; and ii) an additive formulation where each node in the network is modelled by a generalised linear regression model and this formulation applies to observations comprising of binary and/or Gaussian variables, where a logistic link function is used in the former. The additive formulation is analogous to searching for the most appropriate multidimensional Bayesian regression model of the data.

Downloads: