Predict From a `coxnet` Model

Usage

# S3 method for class 'coxnet_model'
predict(
  object,
  new_data,
  type = c("linear_pred", "survival", "time"),
  penalty = NULL,
  eval_time = NULL,
  increasing = TRUE,
  ...
)

Arguments

object: A [coxnet()] model.
new_data: A data frame (or matrix) of new predictors.
type: `"linear_pred"` for the linear predictor, `"survival"` for survival probabilities at `eval_time`, or `"time"` for the restricted mean survival time: the area under the predicted survival curve from 0 to the last time in the training data.
penalty: The penalty to predict at; defaults to the model's `penalty`.
eval_time: For `type = "survival"`, the times to predict survival at.
increasing: For `type = "linear_pred"`: if `TRUE` (the default, as in tidymodels' censored package), the sign is flipped so larger values mean longer survival. Use `FALSE` for glmnet's own sign, where larger values mean higher risk.
...: Not used.

Value

A tibble with one row per row of `new_data`: `.pred_linear_pred`, `.pred_time`, or `.pred`, a list-column of tibbles with `.eval_time` and `.pred_survival`.

Survival is \(S(t \mid x) = \exp(-H_0(t) e^{x^\top \beta})\), with \(H_0\) the Breslow estimate of the cumulative baseline hazard from the training data at `penalty`. Each row's covariates are taken as constant from time 0; to predict along a subject's start/stop covariate path, and score it, use [cv_coxnet()].