Bayesian Edge Screening and Structure Selection for the Ising Model
Bayesian edge screening and structure selection for the Ising model using continuous spike-and-slab prior distributions. A mixture of two normal prior distributions is stipulated on the interaction effects to model edge inclusion and exclusion. A standard normal prior is stipulated on the main effects. Hyperparameters for the normal mixture are automatically determined by fixing the type-1 error. The details of this procedure can be found in Marsman, Huth, Waldorp, and Ntzoufras (https://psyarxiv.com/dg8yx/). The prior distribution on the structures (configurations of edges) is either uniform, or uniform on structure complexity (Beta(1,1)-Binomial). The EM variable selection approach of Ročková and George (Journal of the American Statistical Association, 109(506):828-846, 2014) is used for edge screening and the SSVS approach of George and McCulloch (Journal of the American Statistical Association, 88(423):881-889, 1993) is used for structure selection.
You can install the development version from GitHub with:
install.packages("devtools")
devtools::install_github(“MaartenMarsman/rbinnet”)