Learning a Lattice Graph from Simulated Data

In the following example, a multivariate binary data matrix of 400 observations that will give a 5 x 5 grid graph will be simulated. Our XMRF Ising model is fit to infer the lattice network from simulated data. Models were fitted over a path of 20 regularization values. StARS stability selection with 100 iterations were used to select the optimal network. Figure S3 shows the original network and estimated optimal network plotted with the GMS's default plot function and R image function.

> n = 400
> p = 25
> simdat <- XMRF.Sim(n=n, p=p, model="ISM", graph.type="lattice")
> ismfit <- XMRF(simdat$X, method="ISM", N=100, nlams=20, 
            stability="STAR", th=0.1, beta=0.1)
        
> par(mfrow=c(2,2))
> image(simdat$B)
> image(ismfit$network[[ismfit$opt.index]])
> ml = plotNet(simdat$B, fn="")
> ml = plot(ismfit, fn="", mylayout=ml)

Figure S3: Results of fitting an Ising model to simulated multivariate binary data. The true simulated grid is plotted in (A) and (C). The estimated graph structure via XMRF(...,method="ISM") is plotted in (B) and (D).
Image plot2_46_n400p25th01beta01_a4v2



2015-05-29