modreg.control {dirttee}

Setting fitting values for modreg.

Description

This is an internal function of package dirttee which allows control of the numerical options for fitting mode regression. Typically, users will want to modify the defaults if model fitting is slow or fails to converge.

Usage

modreg.control(
  StartInterval = sqrt(3),
  nStart = 11,
  nInterim = NULL,
  maxit = 100,
  itInterim = 10,
  tol = 10^-4,
  tol_bw_plugin = 10^-3,
  maxit_bw_plugin = 10,
  maxit_penalty_plugin = 10,
  tol_penalty_plugin = 10^-3,
  tol_regopt = tol * 100,
  tol_opt = 10^-3,
  maxit_opt = 200,
  tol_opt2 = 10^-3,
  maxit_opt2 = 200
)

Arguments

StartInterval

Starting values are based on an estimate for the mean and an interval around it. The interval is +-\code{StartInterval} * \sigma. Default is \sqrt{3}.

nStart

Number of starting values, considered in the first iteration. Default is 11.

nInterim

Probably has little impact on speed and result. After itInterim weighted least squares iterations, the number of estimates is reduced from nStart to nInterim estimates. Default is 5.

maxit

Maximum number of iterations for the weighted least squares algorithm. Default is 100.

itInterim

Probably has little impact on speed and result. After itInterim weighted least squares iterations, the number of estimates is reduced from nStart to nInterim estimates. Default is 10.

tol

Convergence criterion for the weighted least squares algorithm. Default is 10^-4.

tol_bw_plugin

Convergence criterion for bandwidth selection in the "Plugin" method. Default is 10^-3.

maxit_bw_plugin

Maximum number of iterations for bandwidth selection in the "Plugin" method. Default is 10.

maxit_penalty_plugin

Maximum number of iterations for penalty selection in the "Plugin" method. Default is 10.

tol_penalty_plugin

Convergence criterion for penalty selection in the "Plugin" method. Default is 10^-3.

tol_regopt

Weighted least squares are recalculated for hyperparameter optimization. This is the convergence criterion within this optimization. Default is tol * 100.

tol_opt

Convergence criterion for the first hyperparameter optimizion. Can be increased to reduce compuation time. Default is 10^-3.

maxit_opt

Maximum number of iterations for the first hyperparameter optimizion. Can be lowered to reduce compuation time. Default is 200.

tol_opt2

Convergence criterion for the second hyperparameter optimizion. Default is 10^-3.

maxit_opt2

Maximum number of iterations for the second hyperparameter optimizion. Default is 200.

Details

The algorithm is described in Seipp et al. (2022). To increase the speed of the algorithm, adapting tol and maxit_opt/maxit_opt2 and other penalty / hyperparameter optimization parameters are a good starting point.

Value

A list with the arguments as components

References

Seipp, A., Uslar, V., Weyhe, D., Timmer, A., & Otto-Sobotka, F. (2022). Flexible Semiparametric Mode Regression for Time-to-Event Data. Manuscript submitted for publication.
Yao, W., & Li, L. (2014). A new regression model: modal linear regression. Scandinavian Journal of Statistics, 41(3), 656-671.