mgm: Structure Estimation for Mixed Graphical Models in high-dimensional Data

We present the R-package mgm for the estimation of mixed graphical models underlying multivariate probability distributions over variables with different domains in high-dimensional data. Our method goes beyond existing methods in that it is the first general method to combine variables with categorical, count-measure and continuous domain, while modeling all variables on their proper domain, which avoids possible loss of information due to transformations. In addition to the presenting the estimation function, we provide a function to sample from pairwise mixed distributions and apply our method to a medical dataset.