[R] Getting runtime error in stepclass

[Douglas Bates]

On 7/5/05, Uwe Ligges <ligges at statistik.uni-dortmund.de> wrote:
> Uwe Ligges wrote:
> 
> > Soma Saha wrote:
> >
> >> Hi!
> >>
> >> I got the following runtime error when I tried to use svm method with
> >> stepclass.
> >>
> >> Error in "colnames<-"(`*tmp*`, value = c("0", "1")) :
> >>         attempt to set colnames on object with less than two dimensions
> >>
> >> I repeated the same sequence of statements but this time I used the
> >> classification function used in the example, i.e., "lda" and it worked
> >> fine but I got the same error when I tried randomForest.
> >>
> >> As the same script worked with a different classification function, I am
> >> wondering if stepclass works with svm and randomForest.
> >>
> >> I using R version 2.0.1 on a Linux machine. I am including the R script I
> >> used below.
> >>
> >> I would be grateful for any suggestion on how to make stepclass work with
> >> these classification functions.
> >>
> >> I would also like to make stepclass work with knn but knn does not have a
> >> 'predict' method, which is a requirement for a classification method to
> >> work with stepclass. Is there any way to get around this?
> >>
> >> Thanks,
> >> Soma
> >>
> >>
> >>
> >> library("e1071")
> >> library("randomForest")
> >> library("klaR")
> >>
> >> td <- read.table("dgdata1.txt", header=TRUE, sep=",")
> >> dgenes <- subset(td, dg == 1, select = dg:eg)
> >> ndgenes <- subset(td, dg == 0, select = dg:eg)
> >> n1 <- nrow(dgenes)
> >> n2 <- nrow(ndgenes)
> >> ndgrows <- 1:n2
> >> selrows <- sample(ndgrows)
> >> sndgenes <- ndgenes[selrows[1:n1],]
> >>
> >> train <- rbind(dgenes, sndgenes)
> >> attach(train)
> >> traind <- subset(train, select = -dg)
> >> trainr <- factor(dg)
> >> detach(train)
> >>
> >> sc_res <- stepclass(traind, trainr, "svm", direction = "forward",
> >> criterion = "AC", fold = 10)
> >
> >
> > Looks like predict.svm does not always return "probabilities" in all
> > cases, hence cannot be used in your case, I guess.
> > svmlight (interface in in packages klaR) will do, though. You also might
> > want to use sknn() (also package klaR), since it has a predict method
> > rather than knn().
> >
> > For randomForest, we will enhance stepclass() to support it in future.
> 
> I just told nonsense, stepclass() does not make sense with
> randomForest(), obviously ... (wonder why nobody shouted?)


Oh, we're just so used to you talking nonsense that we don't bother to
point it out any more :-)