[Rcpp-devel] Grid Search in RcppParallel

[Jeffrey Wong]

I am trying to use RcppParallel to do a fast grid search.  The idea is to
construct a matrix using R's expand.grid, then for each row of that matrix,
x, call f(x).  For simplicity the function f will always return a double.
Since Rcpp can call an R function I was hoping RcppParallel would allow me
to use a parallelFor to process the rows of the grid quickly.  However, it
keeps crashing R, and I am wondering if this is because an R function
cannot be passed to RcppParallel?

require(Rcpp)
require(RcppParallel)

grid = as.matrix(expand.grid(a = 1:5, b = 1:5))

foo = function(pars) {pars[1]^2 + pars[2]^2}

RcppParallelGridSearch(grid, foo)

in .cpp file

#include <Rcpp.h>
using namespace Rcpp;

// [[Rcpp::depends(RcppParallel)]]
#include <RcppParallel.h>
using namespace RcppParallel;

struct GridSearch : public Worker
{
   // source matrix
   const RMatrix<double> grid;
   const Function f;

   // destination vector
   RVector<double> output;

   // initialize with source and destination
   GridSearch(const NumericMatrix grid, const Function f, NumericVector
output)
      : grid(grid), f(f), output(output) {}

   // take the square root of the range of elements requested
   void operator()(std::size_t begin, std::size_t end) {

      for (std::size_t i = begin; i < end; i++) {
        RMatrix<double>::Row parameters = grid.row(i);
        output[i] = as<double>(f(parameters));
      }
   }
};

// [[Rcpp::export]]
NumericVector RcppParallelGridSearch(NumericMatrix grid, Function f) {

  // allocate the output matrix
  NumericVector output(grid.nrow());

  // SquareRoot functor (pass input and output matrixes)
  GridSearch gridSearch(grid, f, output);

  // call parallelFor to do the work
  parallelFor(0, grid.nrow(), gridSearch);

  // return the output matrix
  return output;
}

-- 
Jeffrey Wong
My portfolio: http://jeffreycwong.com <http://jeffreyctwong.com>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/rcpp-devel/attachments/20141004/1cfc0c6c/attachment.html>