Ordering Permutation in Rcpp i.e. base::order()

Question

I have a ton of code using the base::order() command and I am really too lazy to code around that in rcpp. Since Rcpp only supports sort, but not order, I spent 2 m

Answer 1

Another solution based on the C++11:

// [[Rcpp::plugins(cpp11)]]
#include 
using namespace Rcpp;

template 
IntegerVector order_impl(const Vector& x, bool desc) {
    auto n = x.size();
    IntegerVector idx = no_init(n);
    std::iota(idx.begin(), idx.end(), static_cast(1));
    if (desc) {
        auto comparator = [&x](size_t a, size_t b){ return x[a - 1] > x[b - 1]; };
        std::stable_sort(idx.begin(), idx.end(), comparator);
    } else {
        auto comparator = [&x](size_t a, size_t b){ return x[a - 1] < x[b - 1]; };
        std::stable_sort(idx.begin(), idx.end(), comparator);
        // simulate na.last
        size_t nas = 0;
        for (size_t i = 0; i < n; ++i, ++nas)
            if (!Vector::is_na(x[idx[i] - 1])) break;
        std::rotate(idx.begin(), idx.begin() + nas, idx.end());
    }
    return idx;
}

// [[Rcpp::export]]
IntegerVector order2(SEXP x, bool desc = false) {
    switch(TYPEOF(x)) {
    case INTSXP: return order_impl(x, desc);
    case REALSXP: return order_impl(x, desc);
    case STRSXP: return order_impl(x, desc);
    default: stop("Unsupported type.");
    }
}

/***R
int <- sample.int(1000, 1E5, replace = TRUE)
dbl <- runif(1E5)
chr <- sample(letters, 1E5, replace = TRUE)
library(benchr)
benchmark(order(int), order2(int))
benchmark(order(dbl), order2(dbl))
benchmark(order(chr), order2(chr))
*/

Compare performance:

R> int <- sample.int(1000, 1E5, replace = TRUE)

R> dbl <- runif(1E5)

R> chr <- sample(letters, 1E5, replace = TRUE)

R> library(benchr)

R> benchmark(order(int), order2(int))
Benchmark summary:
Time units : microseconds 
        expr n.eval  min lw.qu median mean up.qu  max  total relative
 order(int)    100  442   452    464  482   486 1530  48200      1.0
order2(int)    100 5150  5170   5220 5260  5270 6490 526000     11.2

R> benchmark(order(dbl), order2(dbl))
Benchmark summary:
Time units : milliseconds 
        expr n.eval   min lw.qu median  mean up.qu  max total relative
 order(dbl)    100 13.90 14.00  14.20 14.80  15.8 17.4  1480     1.98
order2(dbl)    100  7.11  7.13   7.15  7.26   7.3  8.8   726     1.00

R> benchmark(order(chr), order2(chr))
Benchmark summary:
Time units : milliseconds 
        expr n.eval   min lw.qu median  mean up.qu   max total relative
 order(chr)    100 128.0 131.0  133.0 133.0 134.0 148.0 13300     7.34
order2(chr)    100  17.7  17.9   18.1  18.2  18.3  22.2  1820     1.00

Note that radix method from the base order much faster.