TempleCBE Custom Loss Functions for survex — survex_loss

Provides survival loss functions implementing TempleCBE's inverse-probability-of-censoring weighted (IPCW) Graf Integrated Brier Score and time-dependent Brier Score for use with survex::model_parts() and survex::model_performance().

Usage

survex_loss_ibs(
  normalization = c("none", "t_max", "survival"),
  max_quantile = 1,
  trunc = 0.05
)

survex_loss_brier(trunc = 0.05)

Arguments

normalization: Optional normalization parameter for survex integral calculation. Can be NULL, "t_max", or "survival".
max_quantile: Upper quantile cutoff for evaluation time window. Default is 1.
trunc: Lower truncation threshold for censoring survival probabilities to prevent inflation by rare tails. Default is 0.05.

Value

A loss function with attributes "loss_type" ("integrated" or "time-dependent") and "loss_name", directly passable to survex.

Examples

if (FALSE) { # \dontrun{
if (requireNamespace("survex", quietly = TRUE)) {
  loss_ibs <- survex_loss_ibs()
  # mp <- survex::model_parts(explainer, loss_function = loss_ibs)
}
} # }