cudaBayesreg (version 0.3-6)

post.shrinkage.mean: Computes shrinkage of fitted estimates over regressions

Description

post.shrinkage.mean computes the mean fitted estimates as a function of the mean regression coefficient estimates over all regressions.

Usage

post.shrinkage.mean(out, X, vreg, plot=T)

Value

a list containingyrecmeanmean of fitted valuesbetamean of estimated coefficients over all regressions

Details

To assess the influence of the hyperparameter $nu$ on the dispersion of the fitted estimates and regression coefficient estimates two plots are available in the package: one in terms of means values, the other in terms of maximum and minimum values. These plots help visualizing shrinkage by analyzing the influence of the hyperparameter $nu$ on the estimates. Different shrinkage plots may be compared for simulations with different $nu$ values.

References

Adelino Ferreira da Silva, A Bayesian Multilevel Model for fMRI Data Analysis, Computer Methods and Programs in Biomedicine, to be published.

See Also

cudaMultireg.slice

Examples

slicedata <- read.fmrislice(fbase="fmri", slice=3, swap=FALSE)
ymaskdata <- premask(slicedata)
fsave1 <- "/tmp/simultest1.sav"
nu1 <- 3
out1 <- cudaMultireg.slice(slicedata, ymaskdata, R=2000, keep=5, nu.e=nu1, fsave=fsave1,
  zprior=FALSE, rng=1 )
fsave2 <- "/tmp/simultest2.sav"
nu2 <- slicedata$nobs
out2 <- cudaMultireg.slice(slicedata, ymaskdata, R=2000, keep=5, nu.e=nu2, fsave=fsave2,
  zprior=FALSE, rng=1 )
vreg <- 2
x1 <- post.shrinkage.mean(out1, slicedata$X, vreg=vreg, plot=F) 
x2 <- post.shrinkage.mean(out2, slicedata$X, vreg=vreg, plot=F) 
par(mfrow=c(1,2), mar=c(4,4,1,1)+0.1)
xlim=range(c(x1$beta, x2$beta))
ylim=range(c(x1$yrecmean, x2$yrecmean))
plot(x1$beta, x1$yrecmean,type="p", pch="+", col="violet", ylim=ylim, xlim=xlim,
xlab=expression(beta), ylab="y")
legend("topright", expression(paste(nu,"=3")), bg="seashell")
plot(x2$beta, x2$yrecmean,type="p", pch="+", col="blue", ylim=ylim, xlim=xlim,
xlab=expression(beta), ylab="y")
legend("topright", expression(paste(nu,"=45")), bg="seashell")
par(mfrow=c(1,1))