improvement of the previous "impute" code :instead of doing the entire loop,index of missing rows have been already registered...

impute2.cpp

#include <Rcpp.h>
using namespace Rcpp;

// [[Rcpp::export]]
List modiframe2(DataFrame& df   ) {
    //nrow and ncol of the dataframe:
    int nrow = df.nrows(), ncol= df.size() ;
    double moy(0);
    //define an empty list~dataframe 
    //List output(ncol)
    List output=clone(df);
    for( int i=0; i<ncol; i++) {
        NumericVector tmp=df[i];
        std::vector<int> missval(nrow);
        int k=0;
        for (int j=0; j<nrow; j++) {
           if ( R_IsNA(tmp[j])==false ) {
              moy=moy+tmp[j];
               missval.pop_back();
              } else {
                missval[k]=j;
              }   
        }
      moy=moy/nrow;
      int msize= missval.size();
      //instead of relooping, I have already stored index of missing rows
      for ( int u=0; u<msize; u++) {
        tmp[missval[u]]=moy;
      }
      output[i]=tmp;
    }
    return wrap(output) ;
}
/*** R
library(microbenchmark)
microbenchmark(modiframe(train), modiframe2(train) )
#Unit: milliseconds
#            expr      min       lq   median       uq      max neval
# modiframe(train) 855.1824 880.3109 993.4610 2060.079 2673.451   100
# modiframe2(train) 621.3642 716.5141 780.8182 1632.077 2285.674   100
*/