Learning Bayesian Networks for Nonparanormal Data

In the literature, structural learning procedures for selecting the directed acyclic graph of a Bayesian network are increasingly explored and specified according to the analyzed data typology. With respect to data drawn from a Gaussian Copula model, the Rank PC algorithm, based on Spearman rank correlation, has been introduced. Moreover, we recently proposed a modified version of the well known Grow-Shrink algorithm, the Copula Grow-Shrink one, based on the Spearman rank correlation and the Copula assumption. Here, we show a simulation study to verify the robustness of our Copula Grow-Shrink algorithm and we discuss the performance results in comparison with the baseline and the Rank PC algorithm.