GT-Transfer

We present an integrated multiple knowledge framework to improve the prevalence estimation for group testing data. Our method built on the public source summary statistics to aviod restriction of experiment setting and data privacy. Moreover, we extend the method to retesting design which can further imporve the estimation ability.

Installation

library(devtools)
install_github("amss-stat/GT-Transfer")

Quick start

library(GTT)
set.seed(4321)

# Simulation data
p=0.01
k=5
m=500
pi_1=0.95
pi_0=0.95
S1=0.999
S0=0.999

d=matrix(rbinom(k*m,size=1,prob=p), nrow=k)
gresult=c()
for (h in 1:m){
 if(sum(d[,h])>0){
   gresult=c(gresult, rbinom(1,1,pi_1))
 }else{
   gresult=c(gresult, rbinom(1,1,1-pi_0))
 }
}
m_1=sum(gresult)
 
m_1i=rep(0,k+1)
retest=which(gresult==1)
R=0
for (r in retest){
 R=0
 for(d_1 in d[,r]){
   if(d_1>0){
     R=R+rbinom(1,1,S1)
   }else{
     R=R+rbinom(1,1,1-S0)
   }
 }
 m_1i[R+1]=m_1i[R+1]+1
}

g_tilde_j=c()
p_tilde_j=c()
v_tilde_j=c()
j=1
while(j<6){
 k_j=sample(c(5,10),size=1)
 m_j=sample(2000:5000,size=1,replace = T)
 pi_1_j=runif(1, min = 0.9, max = 0.999)
 pi_0_j=runif(1, min = 0.9, max = 0.999)
 p_j=p
 g_j=pi_1_j-(pi_0_j+pi_1_j-1)*(1-p_j)^k_j
 x=rbinom(m_j,1,g_j)
 if(sum(x)/m_j > 1-pi_0_j & sum(x)/m_j < pi_1_j){
   p_tilde_j[j]=1-((pi_1_j-sum(x)/m_j)/(pi_0_j+pi_1_j-1))^(1/k_j)
   v_tilde_j[j]=(sum(x)/m_j)*(1-sum(x)/m_j)*(1-p_tilde_j[j])^(2-2*k_j)/((pi_1_j+pi_0_j-1)^2*k_j^2*m_j)
   j=j+1
 }
}

# Estimation
tran = GT_tran(m,k,pi_0,pi_1,m_1,p_tilde_j,v_tilde_j,w = 1,eta = NULL)
print(tran)

tran_retest = GT_tran_retest(m,k,pi_0,pi_1,S0,S1,m_1,m_1i,p_tilde_j,v_tilde_j,w = 1,eta = NULL)
print(tran_retest)