aenetgt-package: Adaptive Elastic Net for Group Testing Data

In gregorkb/aenetgt: Adaptive Elastic Net for Group Testing Data

Adaptive Elastic Net for Group Testing Data

Description

Fits the adaptive elastic net logistic regression estimator for group testing data by using a penalized EM algorithm. The method is described in Gregory, K. B., Wang, D., & McMahan, C. S. (2019) <doi:10.1111/biom.12973>.

Details

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

This package contains functions required to compute the adaptive elastic net logistic regression estimator on group testing data as described in Gregory, Wang, and McMahan (2018). The principal functions are enetgt, which computes the elastic net estimator with a weighted l1 norm, and mlegt, which computes the maximum likelihood estimator. Computing the adaptive elastic net estimator requires obtaining an initial estimate, either from enetgt or mlegt, and using the initial estimator to specify the weights in a second fitting with the enetgt function. The enetgt.grid function fits the elastic net estimator with weighted l1 norm over a grid of tuning parameter values. The aic.bic.eric.enetgt.grid function takes the output of the enetgt.grid function and chooses tuning parameter values according to the AIC, BIC, and ERIC criteria. The functions get.individual.cv.fold.data, get.masterpool.cv.fold.data, get.dorfman.cv.fold.data, and get.assay.cv.fold.data split group testing data into crossvalidation training and testing folds. The function cv.enetgt.grid takes the crossvalidation data folds returned by any of these four functions as well as output from enetgt.grid and chooses the tuning parameter values according to a likelihood-based crossvalidation procedure. The functions individual.assay.gen, masterpool.assay.gen, dorfman.assay.gen, and array.assay.gen are provided for generating group testing data under individual testing, master pool testing, Dorfman testing, and array testing, respectively.

Author(s)

Karl Gregory, Dewei Wang, and Chris McMahan

Maintainer: Karl Gregory <[e-mail]>

References

Gregory, K. B., Wang, D., & McMahan, C. S. (2019). Adaptive elastic net for group testing. Biometrics.

Hui, F.K., Warton, D.I., & Foster, S.D. (2015). Tuning parameter selection for the adaptive lasso using ERIC. Journal of the Americal Statistical Association, 110(509), 262-269.

McMahan, C.S., Tebbs, J.M., Hanson, T.E., & Bilder, C.R. (2017). Bayesian regression for group testing data. Biometrics.

Examples

# generate covariate values and disease statuses for 300 individuals from model0:
data <- model0(300)
X <- data$X
Y.true <- data$Y
b.true <- data$b
b.true

# subject individuals to individual testing
Se <- c(.94) # individual testing sensitivities
Sp <- c(.95) # individual testing specificities
pt.data <- individual.assay.gen(Y.true,Se,Sp,cj=1)
Z <- pt.data$Z
Y <- pt.data$Y

# compute the mle on the group testing data:
mlegt.out <- mlegt(X,Y,Z,Se,Sp)
b.mle <- mlegt.out$b.mle
b.mle

# compute adaptive elastic net estimator over a grid of tuning parameter values
n.lambda <- 8
n.theta <- 2
enetgt.grid.out <- enetgt.grid(X,Y,Z,Se,Sp,n.lambda,n.theta,weights = 1/abs(b.mle[-1]))

# make choices of the tuning parameters according to the aic, bic, and eric criteria
aic.bic.eric.enetgt.grid.out <- aic.bic.eric.enetgt.grid(enetgt.grid.out,Z,X,Se,Sp,"dorfman")
b.aenet.aic <- aic.bic.eric.enetgt.grid.out$b.enet.aic
b.aenet.bic <- aic.bic.eric.enetgt.grid.out$b.enet.bic
b.aenet.eric <- aic.bic.eric.enetgt.grid.out$b.enet.eric
b.aenet.aic 
b.aenet.bic
b.aenet.eric

# make a choice of the tuning parameter using 3-fold crossvalidation:
cv.fold.data <- get.individual.cv.fold.data(X,Y,Z,K=3)
cv.enetgt.grid.out <- cv.enetgt.grid(cv.fold.data,"individual",Se,Sp,enetgt.grid.out$lambda.seq,
						enetgt.grid.out$theta.seq,weights=1/abs(b.mle[-1]),
						B.INIT=enetgt.grid.out$B.ENET)
b.aenet.cv <- enetgt.grid.out$B.ENET[cv.enetgt.grid.out$cv.ind[1],,cv.enetgt.grid.out$cv.ind[2]]
b.aenet.cv

gregorkb/aenetgt documentation built on Oct. 12, 2022, 11:51 a.m.