EMSNM-package: EM Algorithm for Sigmoid Normal Model

In EMSNM: EM Algorithm for Sigmoid Normal Model

Description

It provides a method based on EM algorithm to estimate the parameter of a mixture model, Sigmoid-Normal Model, where the samples come from several normal distributions (also call them subgroups) whose mean is determined by co-variable Z and coefficient alpha while the variance are homogeneous. Meanwhile, the subgroup each item belongs to is determined by co-variables X and coefficient eta through Sigmoid link function which is the extension of Logistic Link function. It uses bootstrap to estimate the standard error of parameters. When sample is indeed separable, removing estimation with abnormal sigma, the estimation of alpha is quite well. I used this method to explore the subgroup structure of HIV patients and it can be used in other domains where exists subgroup structure.

Details

The DESCRIPTION file: This package was not yet installed at build time.

Index: This package was not yet installed at build time.

The EMalgorithm is used to estimate the parameters, EMbootstrap is used to estimate the parameters with bootstrap method. In EMsimulation we can simulate the situation with given parameters, so parameter estimation can be verified.

Author(s)

Linsui Deng <[e-mail]>

Maintainer: Linsui Deng <[e-mail]>

Examples

#parameter initialization
etasize <- 2
classsize <- 2
alphasize <- 3
samplesize <- 100
expriments <- 30
  
etatest <- matrix(c(1,1,
                    0,0),etasize,classsize)

alphatest <- matrix(c(1,0,2,
                      4,3,5),alphasize,classsize)
sigmatest <- 0.5

#test of EMsimulation
EMsimulation_result <- EMsimulation(eta=etatest,alpha=alphatest,sigma=sigmatest,
                            samplesize=samplesize,expriments=expriments,
                            compact_flag=TRUE,C0=5,C1=0.5,C2=5)
index <- which(EMsimulation_result$sigma<0.8)
EMsimulation_result_sort <- EM_result_sort(EMsimulation_result$alpha[index,,],
                                            EMsimulation_result$eta[index,,])
EM_parameter <- EM_parameter_sd(EMsimulation_result_sort$alpha,
                                EMsimulation_result_sort$eta,
                                EMsimulation_result$sigma[index])

#test of EMbootstrap
samplesize <- 1000
X <- matrix(c(matrix(1,samplesize),
                  rnorm(samplesize*(etasize-1))+1),samplesize,etasize)
Z <- matrix(c(matrix(1,samplesize),rbinom(prob=1/2,size=1,n=samplesize),
                 rnorm(samplesize*(alphasize-2))+1),samplesize,alphasize)
                 
Wtest <- Wgenerate(alpha=alphatest,eta=etatest,sigma=sigmatest,X=X,Z=Z)

boots_samplesize <- 100
boots_expriments <- 30
samplesize <- dim(Wtest$X)[1]
EMbootstrap_theta <- EMbootstrap(Wtest$X,Wtest$Y,Wtest$Z,samplesize,
                                 boots_samplesize,boots_expriments,
                                 classsize=2,compact_flag=TRUE,C0=5,C1=0.2,C2=5)
index <- which(EMbootstrap_theta$sigma<0.8)
EMsimulation_result_sort <- EM_result_sort(EMbootstrap_theta$alpha[index,,],
                                             EMbootstrap_theta$eta[index,,])
EM_parameter <- EM_parameter_sd(EMsimulation_result_sort$alpha,
                                EMsimulation_result_sort$eta,
                                 EMbootstrap_theta$sigma[index])                          

EMSNM documentation built on May 2, 2019, 1:41 p.m.