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

In cudaBayesreg: CUDA Parallel Implementation of a Bayesian Multilevel Model for fMRI Data Analysis

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)

Arguments

out	output of MCMC simulation
X	regression matrix used in the simulation
vreg	number of the regression coefficient
plot	{T,F} output plot (default=T)

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.

Value

a list containing

yrecmean	mean of fitted values
beta	mean of estimated coefficients over all regressions

Author(s)

A. Ferreira da Silva, Universidade Nova de Lisboa, Faculdade de Ciencias e Tecnologia,
[e-mail].

See Also

cudaMultireg.slice

Examples

## Not run: 
slicedata <- read.fmrislice(fbase="fmri", slice=3, swap=FALSE)
ymaskdata <- premask(slicedata)
fsave <- paste(tempdir(),"/simultest1",fileext = ".sav", sep="")
nu1 <- 3
out1 <- cudaMultireg.slice(slicedata, ymaskdata, R=2000, keep=5, nu.e=nu1,
  fsave=fsave1, zprior=FALSE, rng=1 )
fsave <- paste(tempdir(),"/simultest2",fileext = ".sav", sep="")
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))

## End(Not run)

cudaBayesreg documentation built on May 29, 2017, 6:19 p.m.