R/bonode.R

In quint: Qualitative Interaction Trees

bonode<-function(Gmat,y,Xmat,tr,dmatrow,dmatsg,parent,parvec,w,nsplit,crit=crit)  {
  #selects best observed parent node
  #create matrix to keep the highest value of the criterion
  nleaf <- ncol(Gmat)
  critmax <- matrix(0,nrow=nleaf,ncol=7)
  critmax[,1] <- 1:nleaf
  dmatsel <- matrix(0,nrow=dim(dmatsg)[1],ncol=nleaf)
  dmatsel <- sapply(1:nleaf,function(kk,dmatrow,dmatsg,nleaf){ ifelse(apply(dmatsg[,c(nleaf,nleaf+1)]==dmatrow[kk],1,sum)!=2,1,0)},nleaf=nleaf,dmatrow=dmatrow,dmatsg=dmatsg)
  critmax[,c(2:7)] <- t(sapply(1:nleaf,function(kk,y,Xmat,tr,Gmat,dmatsg,dmatsel,parent,parvec,w,nsplit,crit){
    bovar(y,Xmat,tr,gm=Gmat[,kk],dmatsg,dmatsel=dmatsel[,kk],parents=parent[-kk,],parvec,w,nsplit,crit=crit) },
    Xmat=Xmat,Gmat=Gmat,y=y,tr=tr,parent=parent,nsplit=nsplit,w=w,parvec=parvec,dmatsg=dmatsg,dmatsel=dmatsel,crit=crit) )
  #which predictor is the best splitting candidate for this split?
  bestrow <- which(critmax[,5]==max(critmax[,5])) [1]
  return(critmax[bestrow,])
}