Learning Mixed Graphical Models

Paper

The full paper can be found at the Arxiv. In this work, we propose a new pairwise graphical model over continuous and discrete variables. This model generalizes both the Gaussian MRF and the usual pairwise discrete MRF. The learned model is sparse by using appropriate group-sparsity regularizers for each edge.

Code

Code for reproducing these results is here and requires UGM and TFOCS. See here for a demo.