[R] (bug?) in survfit in R-2-15-0

[Damjan Krstajic]

Dear all,

 

I am using glmnet + survival and due to the latest
release of glmnet

1.7.4 I was forced to use the latest version of R 2.15.0.
My previous version of R was 2.10.1. I changed glmnet version and R
version and when I started to get weird results I was not sure where the bug was.

 

After putting glmnet back to previous version, I have
found that the bug or weird behaviour happens in survival survfit. My script is
below in email and it is uses glmnet 1.7.3. I get two completely different
answers in different versions of R and in my opinion the older version of
survival package returns correct value.

 

in R 2-10-1 I get

 >
cat(dim(sfit$surv))

 

 >
cat(length(sfit$surv))

32

 >

 

in R-2-15-0 I get

cat(dim(sfit$surv))

62 99

 >
cat(length(sfit$surv))

6138

 >

 Kind regardsDK


library(survival)

library(glmnet)

load(system.file("doc","VignetteExample.rdata",package="glmnet"))

attach(patient.data)

trainX     
<-x[-1,]

trainTime  
<-time[-1]

trainStatus <- status[-1]

fit <-

glmnet(trainX,Surv(trainTime,trainStatus),family="cox",alpha=0.5,maxit=10000)

max.dev.index    
<- which.max(fit$dev.ratio)

optimal.lambda <- fit$lambda[max.dev.index]
optimal.beta  <-
fit$beta[,max.dev.index] nonzero.coef <- abs(optimal.beta)>0 selectedBeta
<- optimal.beta[nonzero.coef]

selectedTrainX  
<- trainX[,nonzero.coef]

coxph.model<- coxph(Surv(trainTime,trainStatus)~

selectedTrainX,init=selectedBeta,iter=0)

selectedTestX <- x[1,nonzero.coef]

sfit<- survfit(coxph.model,newdata=selectedTestX)

cat("\ndim(sfit$surv)")

cat(dim(sfit$surv))

cat("\nlength(sfit$surv)")

cat(length(sfit$surv))