We consider the problem of approximating an underlying distribution by one derived from a dependence polytree. We propose a formal and systematic algorithm, which traverses the undirected tree obtained by the Chow method [IEEE Trans. Inform. Theory 14 (1968) 462], and which subsequently processes the latter using the knowledge of inter-node independence tests. By using the tree structure and these independence tests, our scheme successfully orients the polytree using an application of the depth first search (DFS) strategy to multiple causal basins. The algorithm has been formally proven, and rigorously tested for synthetic and real-life data.

Additional Metadata
Persistent URL dx.doi.org/10.1016/j.ins.2004.01.001
Journal Information Sciences
Ouerd, M. (M.), Oommen, J, & Matwin, S. (S.). (2004). A formal approach to using data distributions for building causal polytree structures. Information Sciences, 168(1-4), 111–132. doi:10.1016/j.ins.2004.01.001