autoselect.shash: Automatic model selection for SinH-ArcSinH continuous norming via BIC

Description

Selects the polynomial degrees for the four components of a cnorm.shash model ($\mu$, $\sigma$, $\epsilon$, $\delta$) by minimizing BIC over a full grid.

Usage

autoselect.shash(
  age,
  score,
  weights = NULL,
  max_mu = 4,
  max_sigma = 3,
  max_epsilon = 3,
  max_delta = 1,
  min_mu = 1,
  min_sigma = 1,
  min_epsilon = 1,
  min_delta = 0,
  delta = 1,
  control = NULL,
  scale = "T",
  parallel = TRUE,
  n_cores = NULL,
  plot = TRUE,
  verbose = TRUE
)

Value

The selected fitted cnormShash model with an additional element $selection containing:

evaluated: data frame of every combination tried, sorted by BIC.
selected: list with the chosen degrees and BIC.

Arguments

age, score: Numeric vectors of predictor and response values.
weights: Optional numeric vector of observation weights.
max_mu, max_sigma, max_epsilon, max_delta: Maximum polynomial degrees for the four shash parameters. Defaults: 4, 3, 3, 1.
min_mu, min_sigma, min_epsilon: Minimum polynomial degrees (default 1).
min_delta: Minimum degree for $\delta$. 0 (default) means the fixed-delta variant is included in the search.
delta: Value of $\delta$ used whenever the candidate uses delta_degree = 0 (i.e. fixed delta). Default 1.
control: Optional control list passed to optim.
scale: Norm scale (default "T").
parallel: Logical; attempt parallel execution. Default TRUE.
n_cores: Number of cores. Defaults to all logical cores.
plot: Logical; plot the selected model. Default TRUE.
verbose: Logical; print progress. Default TRUE.

Details

A degree of 0 for $\delta$ is interpreted as a **fixed** $\delta$ (passed to cnorm.shash as delta_degree = NULL) with value taken from the delta argument. This allows the grid search to choose between a fixed and a varying tail-weight model.

Parallel execution is attempted by default. If the workers cannot access the cNORM namespace, the function transparently falls back to sequential execution.

Examples

Run this code

if (FALSE) {
m <- autoselect.shash(elfe$group, elfe$raw)
m$selection$evaluated
summary(m, age = elfe$group, score = elfe$raw)
}