ref_spec

Usage

ref_spec(
  family = ref_gaussian(),
  location = ~1,
  scale = ~1,
  skew = ~1,
  tail = ~1,
  transform = "identity",
  method = "REML",
  use_bam = TRUE,
  bam_min_n = 20000L,
  control = list()
)

Specification of a distributional reference model

Arguments

Argument	Description
family	A ref_family such as ref_gaussian() or ref_shash().
location, scale, skew, tail	One-sided formulas for the additive predictors. Unused shape formulas default to ~ 1.
transform	Monotone response transform applied before fitting: "identity" (the default), "sqrt", "log", or a Box-Cox power $\lambda \ge 0$. The model is fitted to $h_\lambda(y)=(y^\lambda-1)/\lambda$ ($\log y$ at $\lambda=0$) and every predictive distribution is pushed back to the response scale, so densities carry the log Jacobian and quantiles, medians, hilo(), and CRPS are in the units of y. Any transform other than the identity requires a positive outcome.
method	mgcv smoothness-selection method.
use_bam	Use mgcv::bam() when the family supports it and n >= bam_min_n.
bam_min_n	Minimum rows before bam() is considered.
control	Extra engine controls (list).

Value

An object of class ref_spec.

Examples

ref_spec(family = ref_gaussian(), location = ~ s(age, k = 8) + sex)

<ref_spec> gaussian via mgcv

location: ~s(age, k = 8) + sex

scale: ~1

ref_spec(family = ref_shash(), location = ~ s(age, k = 8), transform = "log")

<ref_spec> shash via mgcv

transform: log

location: ~s(age, k = 8)

scale: ~1

skew: ~1

tail: ~1