greg: Reguarlization graphical model estimation

In statcodes/pgraph: Build Dependency Graphs using Projection

Description

greg calculate regularization graphical model

Usage

greg(z, A, eps = 1e-15, rholist = NULL, gamma = 0.5, trace = FALSE)

Arguments

z	n * p dimensional matrix
A	p * p true graph
eps	a tolerence level for thresholding
gamma	the adaptive lasso penalty parameter
trace	whether to trace to estimation process.

Value

a list.

roc.lasso	roc results for lasso
roc.scad	roc results for scad
roc.alasso	roc results for adaptive lasso

Examples

p=30;
n=120;
tmp=runif(p-1,1,3)
s=c(0,cumsum(tmp));
s1=matrix(s,p,p)
cov.mat.true=exp(-abs(s1-t(s1)))
prec.mat.true=solve(cov.mat.true);
a=matrix(rnorm(p*n),n,p)
data.sa=a%*%chol(cov.mat.true);
true.graph = outer(1:p,1:p,f<-function(x,y){(abs(x-y)==1)})
fit = greg(data.sa, true.graph)
plot(fit$roc.lasso[,1],fit$roc.lasso[,2],type='l')
lines(fit$roc.alasso[,1],fit$roc.alasso[,2],col=2)
lines(fit$roc.scad[,1],fit$roc.scad[,2],col=3)
methodlist = c("ridge","lasso","sam")
fit = vector(mode="list", length=3)
info = vector(mode="list", length=3)
auc = NULL
#plot.new()
for(i in 1:3){
method = methodlist[i]
fit[[i]] = pgraph(data.sa, method = method)
info[[i]] = roc(fit[[i]]$statmat.pearson, true.graph)
auc[i] = sum(-diff(info[[i]]$fpr)*info[[i]]$tpr[-1])
lines(info[[i]]$fpr,info[[i]]$tpr, xlab='FPR',ylab='TPR', type='s', col=i+3)
cat(method, ': auc=', auc[i],'\n')
}
auc

statcodes/pgraph documentation built on May 30, 2019, 9:42 a.m.