lpcr.backward: Conduct longitudinal principal component regression with...

In seonjoo/Lpredict: Prediction with high-dimensional longitudinal predictors using longitudinal principal component analysis

Description

Conduct longitudinal principal component regression with model selection

Usage

lpcr.backward(Y, Xmat = NULL, lfpca = NULL, cov = NULL, T = NULL,
  J = NULL, I = NULL, visit = NULL, varthresh = 0.85,
  timeadjust = FALSE, nfold = 10, lambdalist = 2^(c(-10:10)/2),
  penalty.factor = NULL, M = NULL, method = "AUC", seednum = 1234)

Arguments

Y	Outcome in PCR
Xmat	Longitudinal Predictor Matrix
lfpca	(default=NULL) LFPC results. If it is not specified, the following parameters have to be specified to run LPCA
cov	Covariate
T	Time of theimage collection
J	Total number of observations
I	Total number of subjects
visit	Vector of number of visits per subjects
varthresh	(default=0.99) Threshold for variance explained for both subject-specific and subject-visit specific compoents for dimension selection
timeadjust	(default=TRUE) Scale time per subject
method	(default=AUC) backward selection criteria. c('AUC','bic','aic')
verbose	(default=FALSE)
Nx	Dimension of the subject-specific components
Nw	Dimension of the subject-visit specific components
projectthresh	Threshold for variance explain in the first step of SVD

Value

xi

phix0

phix1

zeta

phiw

Author(s)

Seonjoo Lee, [e-mail]

References

TBA

Examples

I=200
visit=rpois(I,1)+3
time = unlist(lapply(visit, function(x) scale(c(0,cumsum(rpois(x-1,1)+1)))))
J = sum(visit)
V=1000
phix0 = matrix(0,V,3);phix0[1:50,1]<-.1;phix0[1:50 + 50,2]<-.1;phix0[1:50 + 100,3]<-.1
phix1 = matrix(0,V,3);phix1[1:50+150,1]<-.1;phix1[1:50 + 200,2]<-.1;phix1[1:50 + 250,3]<-.1
phiw = matrix(0,V,3);phiw[1:50+300,1]<-.1;phiw[1:50 + 350,2]<-.1;phiw[1:50 + 400,3]<-.1
xi = t(matrix(rnorm(I*3),ncol=I)*c(8,4,2))*3
zeta = t(matrix(rnorm(J*3),ncol=J)*c(8,4,2))*2
Xmat = phix0%*% t(xi[rep(1:I, visit),]) + phix1%*% t(time * xi[rep(1:I, visit),]) + phiw %*% t(zeta) + matrix(rnorm(V*J,0,.1),V,J)
beta=c(1,-1,1)/10
p=exp( - xi %*% beta)/(1 + exp( - xi %*% beta))
Y = unlist(lapply(p,function(x)rbinom(1,1,prob=x)))
re<-hd_lfpca(Ydat,Xmat,T,J,I,visit, varthresh=0.85, timeadjust=FALSE)
lpcr.backward(Y,lfpca=re)

seonjoo/Lpredict documentation built on May 29, 2019, 6:54 p.m.