Create a bootstrap configuration

Groups bootstrap settings into a single object that can be passed to ineqx via the se argument, or to plot methods via the ci argument. All model arguments (data, formulas, etc.) are inherited from the ineqx() call automatically.

Usage

boot_config(
  B = 100L,
  parallel = FALSE,
  ncores = NULL,
  seed = NULL,
  verbose = TRUE,
  cl_type = NULL
)

Arguments

B

Integer, number of bootstrap replicates. Default 100.

parallel

Logical, use parallel computation. Default FALSE.

ncores

Integer, number of cores for parallel. Default: all but one.

seed

Integer, random seed for reproducibility. Default NULL.

verbose

Logical, print progress messages. Default TRUE.

cl_type

Character, parallel cluster type: "fork" (Unix/macOS; workers share memory copy-on-write, so the data is not duplicated per core – faster and far lighter on memory for large data) or "psock" (Windows-compatible; each worker gets its own data copy). Default NULL selects "fork" off Windows and "psock" on Windows; requesting "fork" on Windows warns and falls back to PSOCK.

Value

An object of class "ineqx_boot_config"

Two-stage workflow

For one decomposition, one bootstrap, the single-stage flow via ineqx(..., se = boot_config(...)) or bootstrap_se() is the right choice. If you instead want bootstrap SEs for several decomposition views (e.g. different ystat, different ref) of the same GAMLSS fit, use the two-stage flow: bootstrap_params() caches the resampled params once, then decompose_boot_params() produces SEs per view cheaply.

Examples

if (FALSE) { # \dontrun{
# Bootstrap SEs with 500 replicates
ineqx("income", treat = "treat", group = "group", data = mydata,
      formula_mu = ~ treat * group,
      formula_sigma = ~ treat * group,
      se = boot_config(B = 500, seed = 42))

# Bootstrap CIs in plot (explicit config or "boot" shorthand)
plot(desc_result, ci = boot_config(B = 100))
plot(desc_result, ci = "boot")  # equivalent to boot_config()
} # }