R/varplot.maboost.R

In maboost: Binary and Multiclass Boosting Algorithms

"varplot.maboost" <-
  function(x,plot.it=TRUE,type=c("none","scores"),max.var.show=30,...){
    if(class(x)!="maboost"){
      stop("Object must be of type 'maboost'")
    }
    if(missing(type)){
      type="none"
    }
    iter<-x$iter
    nm<-x$names
    vec<-rep(0,length(nm))
    p=x$dim[2]
    Ctree=x$Ctree;
    if(!Ctree){
    g1<-function(i,obj){
      if(dim(obj[[i]]$frame)[1]<2){
        return(rep(0,p))
      }
      imp<-obj[[i]]$splits[,3]^2
      vals<-match(row.names(obj[[i]]$splits),nm)
      vec=rep(0,p)
      vec[vals]<-imp
      vec
     
    }
  }else{
    g1<-function(i,obj){
      
      vec_t=C5imp(obj[[i]])/100;
      var_names=row.names(vec_t);
      vals<-match(var_names,nm)
      
      vec=rep(0,p)
      vec[vals]<-as.numeric(as.matrix(vec_t));
      
      vec
    }
    
  }
    vec<-1/iter*sqrt(apply(sapply(1:iter,function(i)g1(i,x$model$trees)),1,sum))
    
    vars<- order(vec,decreasing=TRUE)
    n<-length(vec)
    max.v=max.var.show
    if(p<max.v)
      max.v=p
    if(plot.it==TRUE){
      dotchart(vec[vars[max.v:1]],labels=nm[vars[max.v:1]],xlab="Score",main="Variable Importance Plot")
    }
    if(type=="scores"){
      vars=vars[1:max.v]
      t1<-vec[vars]
      attr(t1,"names")<-nm[vars]
      return(t1)
    }
  }