Tuning parameters for the fitting algorithms

Tuning parameters for the fitting algorithms

Usage

pcdreg_control(
  maxit = 2000L,
  reltol = 1e-07,
  accelerate = TRUE,
  profile_maxit = 20000L,
  profile_reltol = 1e-06,
  profile_h = NULL
)

Arguments

maxit

Maximum number of iterations. For the rate model these are EM iterations; for the means model, Newton iterations of the estimating equation, which converges in a handful.

reltol

Convergence tolerance on the largest relative change in the coefficients, and for the rate model in the baseline jump sizes too.

accelerate

Rate model only: whether to extrapolate the EM iterations by the SQUAREM scheme of Varadhan and Roland (2008). This reaches the same fixed point in far fewer passes; set it to FALSE to run plain EM.

profile_maxit, profile_reltol

Limits for the inner runs that maximise over the baseline with the coefficients held fixed, used only by the profile likelihood covariance. These can be looser than reltol without loss: the profile log likelihood is stationary in the baseline at its maximum, so an error of size \(\epsilon\) in the inner solution moves it by only \(O(\epsilon^2)\).

profile_h

Step size for the numerical derivative of the profile log likelihood. The default, NULL, uses \(5 n^{-1/2}\) as recommended by Zeng et al. (2016) and used in the paper.

Value

A list of control values.

References

Varadhan, R. and Roland, C. (2008). Simple and globally convergent methods for accelerating the convergence of any EM algorithm. Scandinavian Journal of Statistics 35, 335–353.

Examples

pcdreg_control(reltol = 1e-9)
#> $maxit
#> [1] 2000
#> 
#> $reltol
#> [1] 1e-09
#> 
#> $accelerate
#> [1] TRUE
#> 
#> $profile_maxit
#> [1] 20000
#> 
#> $profile_reltol
#> [1] 1e-06
#> 
#> $profile_h
#> NULL
#>