cv.alpha: Cross-validation runs for risk predition at a single value of...

In Syksy/ePCR: Ensemble Penalized Cox Regression for Survival Prediction

Cross-validation runs for risk predition at a single value of alpha

Description

Run n-fold cross-validation for a chosen prediction metric at a single value of the L1/L2 norm alpha. A suitable lambda sequence is determined by glmnet, and the cross-validation returns a prediction matrix over the folds over various lambda. This function is mostly called by the higher hierarchy functions, such as cv.grid, which allows varying also the alpha-parameter.

Usage

cv.alpha(
  x,
  y,
  folds = 10,
  alpha = 0.5,
  nlamb = 100,
  verb = 0,
  scorefunc,
  plot = FALSE
)

Arguments

x	The data matrix to use for predictions
y	The response for coxnet; preferably a preconstructed Surv-object
folds	Number of cross-validation folds
alpha	Chosen L1/L2 norm parameter lambda
nlamb	Number of lambda values
verb	Integer indicating level of verbosity, where 0 is silent and 1 provides additional information
scorefunc	Chosen scoring function, e.g. score.cindex or score.iAUC
plot	Should a CV-performance curve be plotted as a function of lambda, indicating min/max/mean/median of CV performance over the folds

Value

A matrix of cross-validation scores, where rows correspond to CV folds and columns to various lambda values chosen by glmnet

Examples

data(TYKSSIMU)
library(survival)
ydat <- Surv(event = yMEDISIMU[,"DEATH"], time = yMEDISIMU[,"LKADT_P"])
set.seed(1)
cvs <- cv.alpha(x = xMEDISIMU, y = ydat, alpha = 0.5, folds = 5, 
	nlamb = 50, verb = 1, scorefunc = score.cindex, plot = TRUE)
cvs

Syksy/ePCR documentation built on Feb. 20, 2024, 10:16 p.m.