Expectation maximization - Beta-Binomial and Uniform Distributions

This function calculates the log-likelihood using the expectation-maximization algorithm with Nelder-Mead numerical optimization and beta distribution with one uniform mixture.

Usage

emstepBBU(parmlist, xm, niter, epsilon, trunc, type = "free")

Arguments

parmlist

A list containing initial alpha, mean, and variance values.

xm

Matrix where the first column is total coverage and the second is the count of base A or B.

niter

Max number of iterates.

epsilon

Epsilon value for convergence tolerance. When the absolute delta log-likelihood is below this value, convergence is reached.

trunc

List of two values representing the lower and upper bounds, \(c_{L}\) and \(c_{U}\).

type

String indicating model type. Options: "free" (estimated parameter(s): alpha, mean, and variance), "fixed" (estimated parameter(s): alpha), "fixed-2" (estimated parameter(s): alpha and variance), or "fixed-3" (estimated parameter(s): variance). If avec is length of 1, fixed and fixed-3 will not be able to return a log-likelihood.

Value

List of elements including the log likelihood, the negative log likelihood, the number of iterates, and the optimized parameter values.

Examples

 if(exists("crazy")){
  p = list(avec = c(0.11, 0.22, 0.34, 0.22, 0.11),
          mvec = c(0.20, 0.33, 0.50, 0.67, 0.80),
          svec = c(0.01, 0.01, 0.01, 0.01, 0.01));
  mout <- emstepBBU(p,
                    xm,
                    niter = 100,
                    epsilon = 0.1,
                    trunc = c(0.0,0.0))
}