hybrid.em.pav: Hybrid EM-PAV Algorithm

In lpetito/groupedCS: Simulations to supplement the paper "Misclassified Grouped Current Status Data"

Description

The function hybrid.em.pav executes the hybrid expectation-maximization pool-adjacent-violators algorithm given in the paper "Misclassified Group Tested Current Status Data." The function expectation computes the "E" step in the EM-PAV hybrid algorithm. The iterations are done by the hybrid.em.pav function.

Usage

hybrid.em.pav(data, initial.p, threshold=0.01, alpha=1, beta=1)
expectation(p, delta, alpha=1, beta=1)

Arguments

data	This is a dataset as generated by the function grouped_csdata_weibull.R
initial.p	These are the initial values for the algorithm for each individual in the dataset. Can come from an external source, or should indicate which column of data contains the initial values.
threshold	Point at which algorithm should stop. Is the sum of the squared distances between last iteration and current iteration. Default is 0.01.
p	Probabilities to be updated by the expectation function. This is a vector the length of the group.
delta	A scalar indicating the group test result.
alpha	Sensitivity. Default is 1.
beta	Specificity. Default is 1.

Details

This is an implementation of the EM algorithm for misclassified grouped current status data, using the pool-adjacent-violators algorithm (Ayers et al., 1955) as the maximization step.

Value

results	These should be treated as the right-continuous step function values (like the x value in the output of gpava from the Isotone package.)
num.iterations	The total number of iterations until convergence.
diff	The sum of the squared differences between the final iteration and the previous iteration. This is always less than the specified threshold.

Author(s)

Lucia Petito

Examples

data <- gen.data.weibull.unif(100, 2, 4, 25, 25, 1, 1)
out <- hybrid.em.pav(data, data$initial.p, alpha=0.9, beta=0.9)
head(out$results)
out$num.iterations
out$diff

lpetito/groupedCS documentation built on May 21, 2019, 7:51 a.m.