R/bga.jackknife.R

In made4: Multivariate analysis of microarray data using ADE4

"bga.jackknife" <-
function(data, classvec,  ...){
	ntrain<-ncol(data)  # Microarray data, samples in columns
        if (!ntrain ==length(classvec)) stop("ncol in training data not equal to length classvec")
        classvec<-checkfac(classvec)             
	nclasses=length(levels(classvec))
	out<-matrix(NA, nrow=ntrain, ncol=(nclasses+2))
        
        jackset<-function(ntrain=10,i=1) {
                # Jackknife, One Leave out analysis
        	# to run use:: for (i in c(1:ntrain)) print(jackset(ntrain,i))
                # ntrain is the ncol in the data, i is the sample to be left out

                ind<-c(1:ntrain)
                newtrain<-ind[-(i)]
                newtest<-ind[i]
                return(list(train=newtrain,test=newtest))
	        }               

	for (i in c(1:ntrain)) {
		ind<-jackset(ntrain,i)
		traindata<-data[,ind$train]
		testdata<- as.matrix(data[,ind$test])
		trainvec<- classvec[ind$train]
		testvec <- classvec[ind$test]
                bga.res <- bga.suppl(traindata, testdata, trainvec, testvec, suponly=TRUE)
 		out[i,]<-as.matrix(bga.res)
	  	}
        colnames(out)<-colnames(bga.res)
    
        
        out[,"closest.centre" ]<-as.matrix(factor(out[,"closest.centre" ], labels=levels(classvec)))
        out[,"predicted"]<-as.matrix(factor(out[,"predicted"], labels=levels(classvec)))
	rownames(out)<-colnames(data)

        # Do some v simple stats
        n.correct<-length(which(out[,"predicted"]== out[,"true.class"]))
        p.correct<-n.correct/ntrain*100
        n.incorrect<-length(which(out[,"predicted"]!= out[,"true.class"]))
        stats=c("No.correct"=n.correct, "No.incorrect"=n.incorrect, "%correct"=round(p.correct,2))
        # return 
	return(list("results"=out, "summary"=stats))
	}