Bootstrap Variable Importance via Selection Frequency — variable

Performs bootstrap resampling and refits the CoxKL elastic-net/LASSO CV procedure B times, then summarizes each variable's selection frequency (proportion of times the variable is selected with nonzero coefficient in the best model).

Usage

variable_importance(
  z,
  delta,
  time,
  stratum = NULL,
  RS = NULL,
  beta = NULL,
  etas,
  B = 10,
  nonzero_tol = 1e-10,
  seed = NULL,
  message = FALSE,
  ncores = 1,
  ...
)

Arguments

z: Numeric covariate matrix/data.frame (n x p). If a data.frame is provided, it will be converted to a numeric matrix via as.matrix(z).
delta: Numeric vector of event indicators.
time: Numeric vector of observed times.
stratum: Optional stratum vector. Default NULL.
RS: Optional external risk scores. Default NULL.
beta: Optional external coefficients. Default NULL.
etas: Numeric vector of candidate eta values.
B: Integer. Number of bootstrap replications.
nonzero_tol: Numeric tolerance for defining "selected". Default 1e-10.
seed: Optional integer seed for reproducibility.
message: Logical. Whether to print progress messages. Default FALSE.
ncores: Integer. Number of parallel cores. Default 1 (sequential execution).
...: Additional arguments passed to cv.coxkl_enet() (e.g., alpha, lambda, nlambda, lambda.min.ratio, nfolds, cv.criteria, c_index_stratum, etc.).

Value

An object of class "variable_importance" with fields:

freq: Named numeric vector of selection frequencies (length p).
count: Named integer vector of selection counts (length p).
B: Number of bootstrap replications.
call: Matched call.