theta.control: Control arguments for theta update algorithm

In gldrm: Generalized Linear Density Ratio Models

Description

This function returns control arguments for the θ update algorithm. Each argument has a default value, which will be used unless a different value is provided by the user.

Usage

theta.control(eps = 1e-10, maxiter = 100, maxhalf = 20, maxtheta = 500,
  logit = TRUE, logsumexp = FALSE)

Arguments

eps	Convergence threshold for theta updates. Convergence is evaluated separately for each observation. An observation has converged when the difference between b'(θ) and μ is less than epsTheta.
maxiter	Maximum number of iterations.
maxhalf	Maximum number of half steps allowed per iteration if the convergence criterion does not improve.
maxtheta	Absolute value of theta is not allowed to exceed maxtheta.
logit	Logical for whether logit transformation should be used. Use of this stabilizing transformation appears to be faster in general. Default is TRUE.
logsumexp	Logical argument for whether log-sum-exp trick should be used. This may improve numerical stability at the expense of computational time.

Value

Object of S3 class "thetaControl", which is a list of control arguments.

gldrm documentation built on May 2, 2019, 12:59 p.m.