SAMM (version 1.1.1)

loglikfuncmmmkmv: Calculate the loglikeligood for a general mixed model

Description

Calculate the loglikeligood for the LMM model expressed as $$Y=XB+\sum_{j=1}^k Z_j G_j+WE,$$ where \(Y\) is the \(n \times d\) response variable, \(X\) is the \(n \times q\) design matrix of \(q \times d\) the fixed effects \(B\), \(Z_j\) for \(j=1,2,\ldots,k\) (\(k\geq 1\)) are the \(n \times q_j\) design matrices of the \(q_j \times d\) random effects \(G_j,\) and \(W\) is the \(n \times t\) design matrix of \(t \times d\) residual effecs \(E\). The random effects and the residual are independently distributed, and have matrix variate distributions (\(G_j\sim N_{q_j \times d}(0_{q_j \times d}, K_j,\Sigma_j)\) for \(j=1,2,\ldots,k\) and \(E\sim N_{t \times d}(0_{t \times d}, R,\Sigma_E)\)).

Usage

loglikfuncmmmkmv(Y,X,Zlist, Klist, sigmahatlist, B,W,R )

Arguments

Y

a numeric vector for the parameters (a mapping of the original parameters)

X

a numeric matrix (see examples and details)

Zlist

a numeric matrix (see examples and details)

Klist

a numeric matrix (see examples and details)

sigmahatlist

a numeric matrix (see examples and details)

B

bla bla

W

bla bal

R

bla bla

Value

bla bla

Details

bla bla

Examples

# NOT RUN {
library(SAMM)
n=100
nsample=80
rhotrans=5
ar1cov_cppforR(c(rhotrans),matrix(5))
rho=(2/pi)*atan(rhotrans)
rho
tan((pi/2)*(rho))

M1<-matrix(rbinom(n*300, 2, .2)-1, nrow=n)
K1<-relmatcov_cppforR(c(.01), M1)

M2<-matrix(rbinom(n*300, 2, .2)-1, nrow=n)
K2<-relmatcov_cppforR(c(0.03), M2)
W=(diag(5)[sample(1:5,n, replace=TRUE),])
covY<-3*K1+5*K2+10*(W%*%ar1cov_cppforR(c(rhotrans),matrix(5))%*%t(W))
K1[1:5,1:5]
dim(W)
dim(ar1cov_cppforR(c(6),matrix(5)))
Y<-10+crossprod(chol(covY),rnorm(n))


#training set
Trainset<-sample(1:n,nsample)
ytrain=Y[Trainset]
Xtrain=matrix(rep(1, n)[Trainset], ncol=1)
Ztrain=diag(n)[Trainset,]
Wtrain=W[Trainset,]

samout<-SAMM(Y=matrix(ytrain,ncol=1),X=Xtrain,
Zlist=list(Ztrain, Ztrain), Klist=list(K1,K2),
lambda=0, W=Wtrain,R=list("ar1",c(0),matrix(5,1,1)),
Siglist=list("","",""), corfunc=c(F,F,T), corfuncfixed=c(F,F,F),
sigfunc=c(F,F,F),mmalg="dermm_reml2", tolparconv=1e-10,
tolparinv=1e-10,maxiter=1000,geterrors=F)
samout$corfuncparamslist[[3]]
rhohat=(2/pi)*atan(samout$corfuncparamslist[[3]])
rhohat
ar1cov_cppforR(c(samout$corfuncparamslist[[3]]),matrix(5,1,1))
# }