Fig. 3From: Learning mixed graphical models with separate sparsity parameters and stability-based model selectionComparison of edge recovery performance of MGM and SCGGM on continuous-continuous (cc), continuous-discrete (cd) and both edge types. Matthews correlation is averaged over 20 simulated datasets with linear continuous-discrete interactions and 20 datasets with non-linear interactions with error bars ± one standard error. Sparsity parameters for both methods selected by StEPSBack to article page