R分位数使用Rcpp

Question

I'm interested in how to go about coding the default quantile method used by R using Rcpp. There is a basic solution here Rcpp quantile implementation . As noted in the solution it won't work as well for small vectors.

However, I would really like to implement the base R (type 7 method), shown here https://svn.r-project.org/R/trunk/src/library/stats/R/quantile.R

The basic code looks like:

x <- 1:100 # make a test vector
probs <- c(0.05, 0.95)
n <- length)x)

# R code
index <- 1 + (n - 1) * probs
lo <- floor(index)
hi <- ceiling(index)
x <- sort(x, partial = unique(c(lo, hi)))
qs <- x[lo]
i <- which(index > lo)
h <- (index - lo)[i] # > 0  by construction
qs[i] <- (1 - h) * qs[i] + h * x[hi[i]]

which, for example gives:

quantile(1:100, probs = c(0.05, 0.95))
5%   95% 
5.95 95.05 

It would really help to see how you approach this using Rcpp. I use RStudio and am used to running simpler Rcpp code, but I am stuck on this one.

Any help would be much appreciated.

Thanks

David

Answer 1

What about calling the quantile function of R within Rcpp like this?

#include <Rcpp.h>
using namespace Rcpp;

// [[Rcpp::export]]
NumericVector quantileCpp(NumericVector x, NumericVector probs) {
  Environment stats("package:stats");
  Function quantile = stats["quantile"];
  int npr = probs.size();
  NumericVector ans(npr);
  for(int i=0; i<npr; i++){
    ans[i] = as<double>(quantile(x, probs[i]));
  }
return ans;
}

save the code as 'quantileCpp.cpp' and run:

> Rcpp::sourceCpp('quantileCpp.cpp')

> quantileCpp(1:100,c(0.05,0.95))
[1]  5.95 95.05

> quantile(1:100,c(0.05,0.95))
   5%   95% 
 5.95 95.05 

I know is not the fastest solution (in terms of computational time), but is the easiest solution I can think off.