On Sat, Oct 27, 2012 at 7:50 AM, Dirk Eddelbuettel <e...@debian.org> wrote:
>
>
> We make them look like STL containers by giving them begin(),
> end(), ... member functions but they are still R objects ("SEXP") underneath.
>
> That means in particular that the carefully crafted STL algos with their
> performance guarantees will NOT have these guarantees.
Ah. Thanks for the clarification.
> Someone should sit down and write a C++ variant of sample for Rcpp. Having a
> starting point would help, others can then tune.
I spent some time looking at this today. Guidance/advice on the best
approach here?
Assuming we desire results identical to R sample(), the simple
approach is to follow do_sample in src/main/random.c. The functions
we'd use are ProbSampleReplace, ProbSampleNoReplace, SampleReplace,
SampleNoReplace, and FixupProb.
Q1 -- Are there any immediate drawbacks to cloning-and-transposing
these functions straight from R core?
Q2 -- Given the above, is the following pseudo-code a reasonable
approach? I'm uncertain about the templated instantiation of the
return object.
template <class T>
T sample( const T &x, unsigned int size, bool replace ) {
// user's responsibility? // RNGScope scope;
ans = Vector<T>(size);
ans.index = IntegerVector(size);
if ( replace ) {
... SampleReplace ...
} else {
... SampleNoReplace ...
}
for (int ii = 0; ii<size; ii++ ) {
ans[ii] = x[ ans.index[ii] ];
}
}
T sample( const T &x, unsigned int size, bool replace, const
NumericVector &probs ) {
// user's responsibility? // RNGScope scope;
ans = Vector<T>(size);
ans.index = IntegerVector(size);
...FixupProb...
if ( replace ) {
... ProbSampleReplace ...
} else {
... ProbSampleNoReplace ...
}
for (int ii = 0; ii<size; ii++ ) {
ans[ii] = x[ ans.index[ii] ];
}
return ans;
}
-Christian
--
A man, a plan, a cat, a ham, a yak, a yam, a hat, a canal – Panama!
_______________________________________________
Rcpp-devel mailing list
Rcpp-devel@lists.r-forge.r-project.org
https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/rcpp-devel
