init_param: Build an Initial Value List for BayesSIM Models

Description

init_param is a convenience helper that constructs a nested initial value list for a given combination of index vector and link function. It starts from the model-specific default prior, and then overwrites only those components for which the user supplies non-NULL arguments.

This allows users to modify selected hyper-parameters without having to know or manually reconstruct the full nested prior list structure.

Usage

init_param(
  indexprior,
  link,
  index = NULL,
  index_nu = NULL,
  index_psi = NULL,
  index_pi = NULL,
  link_beta = NULL,
  link_k = NULL,
  link_knots = NULL,
  link_lengthscale = NULL,
  link_amp = NULL,
  link_kappa = NULL,
  link_inv_lambda = NULL,
  sigma2 = NULL
)

Value

A nested list with components index, link, and sigma2.

Arguments

indexprior: Character scalar indicating the prior for the index. Typically one of "fisher", "sphere", "polar", or "spike". The valid options mirror those used in the corresponding model functions.
link: Character scalar indicating the link function family. Typically "bspline" for B-spline link functions or "gp" for Gaussian process link functions. The valid options mirror those used in the corresponding model functions.
index, index_nu, index_psi, index_pi: Optional initial values for index and related parameter values.
link_beta, link_k, link_knots, link_lengthscale, link_amp, link_kappa, link_inv_lambda: Optional initial values for components under link functions.
sigma2: Optional numeric scalar giving the initial value of \(\sigma^2\).

Details

init_param(indexprior, link) can be used to obtain the random initial values list for the requested combination of index prior and link function. For any argument that is not NULL, the corresponding field in the nested prior list is overwritten.

The detailed meaning and recommended choices for each initial values depend on the specific model, index vector and link function. For those details, please refer to the documentation of the corresponding model-fitting functions.

Examples

Run this code

## Default initial values for Fisher index + B-spline link:
i0 <- init_param("fisher", "bspline")

## Modify only a few initial values:
i1 <- init_param(
  indexprior = "fisher",
  link       = "bspline",
  index      = c(1, 0, 0),      # initial direction of the index
  link_beta  = rep(0, 21),      # initial values for spline coefficients
  sigma2     = 0.1              # initial value for sigma^2
)

## Example with GP link:
i2 <- init_param(
  indexprior        = "sphere",
  link              = "gp",
  link_lengthscale  = 0.2,      # initial GP length-scale
  link_amp          = 1.5,      # initial GP amplitude
  sigma2            = 1         # initial variance
)