The topic of how to do the equivalent of the last step here
set.seed(42)
a <- sort(round(runif(10)*100))
a[ a %% 2 == 0]
ie how to subset by some boolean vector had come up before, both here and on
StackOverflow.
A while back, I got hold of a used copy of Austern's classic "Generic
Programming and the STL" in which I found (by random luck, I guess) the
remove_copy() function which takes a vector and copies all values different
from a given 'flagged' value. Very nice -- all is missing then is a
preceding call to transform() in its two vector form to flag values given a
second boolean vector. Combine the two and you get this:
#include <Rcpp.h>
using namespace Rcpp;
using namespace std;
const double flagval = __DBL_MIN__; // works
//const double flagval = NA_REAL; // does not
// simple double value 'flagging' function
inline double flag(double a, bool b) { return b ? a : flagval; }
// [[Rcpp::export]]
NumericVector subsetter(NumericVector a, LogicalVector b) {
// We use the flag() function to mark values of 'a'
// for which 'b' is false with the 'flagval'
transform(a.begin(), a.end(), b.begin(), a.begin(), flag);
// We use sugar's sum to compute how many true values to expect
NumericVector res = NumericVector(sum(b));
// And then copy the ones different from flagval from a into
// res using the remove_copy function from the STL
remove_copy(a.begin(), a.end(), res.begin(), flagval);
return res;
}
and with that can do
subsetter(a, a %% 2 == 0)
which (of course) passes all.equal() with the R solution. I also did some
tests for 1e6 elements in the vector and the logical vector --- and this is a
tad quicker than R itself.
This would need some more work for a properly templated solution using traits
to the sentinel values right. The NA_REAL or R_NaN values do unfortunately
not work as they cannot be used with equality tests, so we waste another
value (which picked as __DBL_MIN__).
Dirk
--
Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com