Vector Subsetting in Rcpp

Kevin Ushey — written Mar 16, 2014 — source

Rcpp 0.11.1 has introduced flexible subsetting for Rcpp vectors. Subsetting is implemented for the Rcpp vector types through the [ operator, and intends to mimic R’s [ operator for most cases.

We diverge from R’s subsetting semantics in a few important ways:

For integer and numeric vectors, 0-based indexing is performed, rather than 1-based indexing, for subsets.
We throw an error if an index is out of bounds, rather than returning an NA value,
We require logical subsetting to be with vectors of the same length, thus avoiding bugs that can occur when a logical vector is recycled for a subset operation.

Some examples are showcased below:

#include <Rcpp.h>
using namespace Rcpp;

// [[Rcpp::export]]
NumericVector positives(NumericVector x) {
    return x[x > 0];
}

// [[Rcpp::export]]
List first_three(List x) {
    IntegerVector idx = IntegerVector::create(0, 1, 2);
    return x[idx];
}

// [[Rcpp::export]]
List with_names(List x, CharacterVector y) {
    return x[y];
}

x <- -5:5
positives(x)

[1] 1 2 3 4 5

l <- as.list(1:10)
first_three(l)

[[1]]
[1] 1

[[2]]
[1] 2

[[3]]
[1] 3

l <- setNames(l, letters[1:10])
with_names(l, c("a", "e", "g"))

$a
[1] 1

$e
[1] 5

$g
[1] 7

Most excitingly, the subset mechanism is quite flexible and works well with Rcpp sugar. For example:

#include <Rcpp.h>
using namespace Rcpp;

// [[Rcpp::export]]
NumericVector in_range(NumericVector x, double low, double high) {
    return x[x > low & x < high];
}

// [[Rcpp::export]]
NumericVector no_na(NumericVector x) {
    return x[ !is_na(x) ];
}

bool is_character(SEXP x) {
    return TYPEOF(x) == STRSXP;
}

// [[Rcpp::export]]
List charvecs(List x) {
    return x[ sapply(x, is_character) ];
}

set.seed(123)
x <- rnorm(5)
in_range(x, -1, 1)

[1] -0.56048 -0.23018  0.07051  0.12929

no_na( c(1, 2, NA, 4, NaN, 10) )

[1]  1  2  4 10

l <- list(1, 2, "a", "b", TRUE)
charvecs(l)

[[1]]
[1] "a"

[[2]]
[1] "b"

And, these can be quite fast:

library(microbenchmark)
R_in_range <- function(x, low, high) {
    return(x[x > low & x < high])
}
x <- rnorm(1E5)
identical( R_in_range(x, -1, 1), in_range(x, -1, 1) )

[1] TRUE

microbenchmark( times=5, 
    R_in_range(x, -1, 1),
    in_range(x, -1, 1)
)

Unit: milliseconds
                 expr   min    lq median    uq   max neval
 R_in_range(x, -1, 1) 8.168 8.556   9.02 9.073 9.223     5
   in_range(x, -1, 1) 5.210 5.424   5.48 5.507 6.233     5

R_no_na <- function(x) {
    return( x[!is.na(x)] )
}
x[sample(1E5, 1E4)] <- NA
identical(no_na(x), R_no_na(x))

[1] TRUE

microbenchmark( times=5,
    R_no_na(x),
    no_na(x)
)

Unit: milliseconds
       expr   min    lq median   uq   max neval
 R_no_na(x) 3.958 3.960  4.019 4.02 4.458     5
   no_na(x) 1.891 1.936  1.961 2.02 2.755     5

We hope users of Rcpp will find the new subset semantics fast, flexible, and useful throughout their projects.

tags: vector basics

Extending R with C++ and Fortran — Dirk Eddelbuettel and JBrandon Duck-Mayr
Benchmarking Rcpp code with RcppClock — Zach DeBruine
Simulation Smoother using RcppArmadillo — Tomasz Woźniak

Vector Subsetting in Rcpp

Related Articles