tramvs (version 0.0-8)

abess_tram: Optimal subset selection for transformation models

Description

Optimal subset selection for transformation models

Usage

abess_tram(
  formula,
  data,
  modFUN,
  supp,
  mandatory = NULL,
  k_max = supp,
  thresh = NULL,
  init = TRUE,
  m_max = 10,
  m0 = NULL,
  ...
)

Value

List containing the fitted model via modFUN, active set

A and inactive set I.

Arguments

formula

object of class "formula".

data

data frame containing the variables in the model.

modFUN

function for fitting a transformation model, e.g., BoxCox().

supp

support size of the coefficient vector

mandatory

formula of mandatory covariates, which will always be included and estimated in the model. Note that this also changes the intialization of the active set. The active set is then computed with regards to the model residuals of modFUN(mandatory, ...) instead of the unconditional model.

k_max

maximum support size to consider during the splicing algorithm. Defaults to supp.

thresh

threshold when to stop splicing. Defaults to 0.01 * supp * p * log(log(n)) / n$, where p denotes the number of predictors and n the sample size.

init

initialize active set. Defaults to TRUE and initializes the active set with those covariates that are most correlated with score residuals of an unconditional modFUN(update(formula, . ~ 1)).

m_max

maximum number of iterating the splicing algorithm.

m0

Transformation model for initialization

...

additional arguments supplied to modFUN.

References

Siegfried, S., Kook, L., & Hothorn, T. (2023). Distribution-free location-scale regression. The American Statistician, 77(4), 345-356. tools:::Rd_expr_doi("10.1080/00031305.2023.2203177")

Examples

set.seed(24101968)
library(tramvs)

N <- 1e2
P <- 5
nz <- 3
beta <- rep(c(1, 0), c(nz, P - nz))
X <- matrix(rnorm(N * P), nrow = N, ncol = P)
Y <- 1 + X %*% beta + rnorm(N)

dat <- data.frame(y = Y, x = X)

abess_tram(y ~ ., dat, modFUN = Lm, supp = 3)