SVC_selection_control {varycoef}

SVC Selection Parameters

Description

Function to set up control parameters for SVC_selection. The underlying Gaussian Process-based SVC model is defined in SVC_mle. SVC_selection then jointly selects fixed and random effects of the GP-based SVC model using a penalized maximum likelihood estimation (PMLE). In this function, one can set the parameters for the PMLE and its optimization procedures (Dambon et al., 2022).

Usage

SVC_selection_control(
  IC.type = c("BIC", "cAIC_BW", "cAIC_VB"),
  method = c("grid", "MBO"),
  r.lambda = c(1e-10, 10),
  n.lambda = 10L,
  n.init = 10L,
  n.iter = 10L,
  CD.conv = list(N = 20L, delta = 1e-06, logLik = TRUE),
  hessian = FALSE,
  adaptive = FALSE,
  parallel = NULL,
  optim.args = list()
)

Arguments

IC.type

(character(1))
Select Information Criterion.

method

(character(1))
Select optimization method for lambdas, i.e., shrinkage parameters. Either model-based optimization (MBO, Bischl et al., 2017 <arXiv:1703.03373>) or over grid.

r.lambda

(numeric(2))
Range of lambdas, i.e., shrinkage parameters.

n.lambda

(numeric(1))
If grid method is selected, number of lambdas per side of grid.

n.init

(numeric(1))
If MBO method is selected, number of initial values for surrogate model.

n.iter

(numeric(1))
If MBO method is selected, number of iteration steps of surrogate models.

CD.conv

(list(3))
List containing the convergence conditions, i.e., first entry is the maximum number of iterations, second value is the relative change necessary to stop iteration, third is logical to toggle if relative change in log likelihood (TRUE) or rather the parameters themselves (FALSE) is the criteria for convergence.

hessian

(logical(1))
If TRUE, Hessian will be computed for final model.

adaptive

(logical(1))
If TRUE, adaptive LASSO is executed, i.e., the shrinkage parameter is defined as \lambda_j := \lambda / |\theta_j|.

parallel

(list)
List with arguments for parallelization, see documentation of optimParallel.

optim.args

(list)
List of further arguments of optimParallel, such as the lower bounds.

Value

A list of control parameters for SVC selection.

Author(s)

Jakob Dambon

References

Bischl, B., Richter, J., Bossek, J., Horn, D., Thomas, J., Lang, M. (2017). mlrMBO: A Modular Framework for Model-Based Optimization of Expensive Black-Box Functions, ArXiv preprint https://arxiv.org/abs/1703.03373

Dambon, J. A., Sigrist, F., Furrer, R. (2022). Joint Variable Selection of both Fixed and Random Effects for Gaussian Process-based Spatially Varying Coefficient Models, International Journal of Geographical Information Science doi: 10.1080/13658816.2022.2097684

Examples

# Initializing parameters and switching logLik to FALSE
selection_control <- SVC_selection_control(
  CD.conv = list(N = 20L, delta = 1e-06, logLik = FALSE)
)
# or
selection_control <- SVC_selection_control()
selection_control$CD.conv$logLik <- FALSE