customPenalized: Penalized regression using the 'penalized' R package

In bernau/survHDExtra: Addtitional Features for package survHD

Description

Ridge or Lasso penalized regression. Usually, this function is not called directly but embedded in a call to customSurv from the main package survHD (see examples).

Usage

customPenalized(Xlearn, Ylearn, learnind, ...)

Arguments

Xlearn	Gene expression data (a data.frame)
Ylearn	Survival Response, an object of class Surv.
learnind	An index vector specifying the observations that belong to the learning set. May be missing; in that case, the learning set consists of all observations and predictions are made on the learning set.
...	Further arguments to be passed to function penalized: penalty Type of penalization: "ridge", "lasso", or "elasticnet". unpenalizedvarnames Character vector which is a subset of colnames(X), naming variables which will be unpenalized. ... Additional arguments that shall be passed to downstream functions. Potential downstream functions include optL1, optL2, and penalized from the penalized package, and opt2D from the pensim package. If lambda is not specified in the ..., then a tuning will be performed using penalized:optL1 (lasso), penalized:optL2 (ridge), or pensim:opt2D. For ridge or lasso, lambda (if specified at all) should be a single numeric value. For elasticnet, lambda (if specified at all) should be a numeric vector of length two. Whether or not tuning is performed, regression for the final model is performed using the penalized function from the penalized R package.

Value

An object of class ModelLearned.

Author(s)

Levi Waldron [e-mail] Christoph Bernau [e-mail] Martin Slawski [e-mail] Anne-Laure Boulesteix [e-mail]

References

Goeman, J.J. (2010) L1 Penalized Estimation in the Cox Proportional Hazards Model, Biometrical Journal, 52, 70-84

See Also

learnSurvival, penalized

Examples

##load example data
if(require(pensim)){
require(survHD)
require(survHDExtra)
data(beer.exprs,package="pensim")
data(beer.survival,package="pensim")
##create Surv object (y)
beerY <- with(beer.survival,Surv(os,status))
##create mini X matrix (only using first 200 genes)
beerX <- t(as.matrix(beer.exprs))
beerX <- beerX[,1:200]
colnames(beerX) <- make.names(colnames(beerX),unique=TRUE)
##scale everything for testing.
beerX <- scale(beerX)
##define training and test sets:
set.seed(4)
allind <- 1:nrow(beerX)
learnind <- sample(allind,size=round(length(allind)*9/10))
testind <- allind[-learnind]
##fit a model with ridge regression, and with L2 penalty equal to 100:
fit.penalized <- customSurv(X=beerX,y=beerY, learnind=learnind, penalty="ridge", lambda=100,customSurvModel=customPenalized)
##linear risk score
predict(fit.penalized@model,newdata=beerX[testind,],type='lp')
##survival probabilities at each uncensored time
mytimegrid <- sort(c(0.0,beerY[beerY[,2]==1,1]))
SurvivalProbs <- predict(fit.penalized,beerX[testind,],type='SurvivalProbs',timegrid=mytimegrid,gbm=TRUE)@SurvivalProbs
plot(SurvivalProbs,xlab="Time", ylab="Probability of Survival")
}

bernau/survHDExtra documentation built on May 12, 2019, 4:22 p.m.