Rcpp quantile implementation

问题

I have a Rcpp double containing a NumericVector x. I would like to get the .95 quantile of such x within the Rcpp code flow. I do not know how can I get it. Is there an RcppArmadillo implementation?

回答1:

I'm not a Rcpp expert and my function probably needs lots of improvement, but it seems that you can easily create your own Rcpp quantile function (which will be not so accurate on small vectors due to high chance to skewness and problems with indexing on non-integer indexes, but improves as the vector grows)

library(Rcpp) # You can use sourceCpp() instead of cppFunction if you wish
cppFunction('NumericVector Cquantile(NumericVector x, NumericVector q) {
  NumericVector y = clone(x);
  std::sort(y.begin(), y.end());
  return y[x.size()*(q - 0.000000001)];
}')

Testing on 1e+3 vector

set.seed(123)
y <- rnorm(1000)
qs <- seq(0, 100, 25)/100

quantile(y, qs)
#           0%          25%          50%          75%         100% 
# -2.809774679 -0.628324243  0.009209639  0.664601867  3.241039935 


setNames(Cquantile(y, qs), paste0(qs * 100, "%"))
#          0%         25%         50%         75%        100% 
# -2.80977468 -0.62957874  0.00729009  0.66441586  3.24103993

Looks close enough. So the 95% quantile for our vector y would be

Cquantile(y, .95)
## [1] 1.675697

There are also some Sugar quantile functions for knows distributions such normal, gamma, etc. that can be used instead, see here for some examples.

来源：https://stackoverflow.com/questions/26786078/rcpp-quantile-implementation

标签

rcpp