Understanding passing armadillo matrices to R functions via RInside

Question

I'm trying to use R within C++ via RInside. I'm having trouble passing armadillo matrices to R and returning a result. Below I am able to return a results from an R library function, however I get the wrong result. I'm using the skewness function from the moments package as an example which works as should in R. I checked the examples from RInside and i'm still unsure how to use RcppArmadillo. How do I properly pass an armadillo matrix in c++ to R?

    #include <RInside.h>                   
    #include <RcppArmadillo.h>

    using namespace std;
    using namespace arma;

    int main(int argc, char *argv[]) {
        RInside R(argc, argv);  


        string R_libs = "suppressMessages(library(moments));";

        R.parseEvalQ(R_libs);

        mat A = randu<mat>(5,5);

        R["A"] = A;

        string R_skewness = "B <- skewness(A);";
        //this fails
        mat B = Rcpp::as<mat>(R.parseEval(R_skewness)); //terminate called after throwing an instance of 'Rcpp::not_a_matrix'   

        //this works but wrong
        mat B = Rcpp::as<vec>(R.parseEval(R_skewness)); // returns only 1 number, should be 5 ( 1 for each columnn), same result if i change mat B to vec B
        exit(0);
 }

score 5 · Accepted Answer · edited Nov 14 '12 at 16:08

The way we implemented as<mat> requires that the R object you pass is a matrix. And in your example B is a vector:

> A <- matrix( runif(25), ncol = 5)
> A
           [,1]      [,2]       [,3]       [,4]      [,5]
[1,] 0.19215339 0.5857249 0.14345222 0.32154176 0.6162155
[2,] 0.95753898 0.9618379 0.06239842 0.06200197 0.7044018
[3,] 0.33575790 0.1372804 0.03027635 0.62662467 0.9778451
[4,] 0.16504957 0.1919765 0.49176372 0.94841456 0.2914772
[5,] 0.01570709 0.8055231 0.51218581 0.79562809 0.6939380
> B <- skewness( A )
> B
[1]  1.15196587 -0.04547576  0.32186257 -0.30788111 -0.29251009

For conversion to arma::vec I don't reproduce the behavior you see. The arma::vec has 3 elements:

require( RcppArmadillo )    ## and make sure you have Rcpp 0.10.0 or later

sourceCpp( code = '
// [[Rcpp::depends("RcppArmadillo")]]

#include <RcppArmadillo.h>

using namespace arma ; 
using namespace Rcpp ;

// [[Rcpp::export]]
List foo( NumericVector x){
    vec B = Rcpp::as<vec>(x); 

    return List::create( 
        _["nrows"] = B.n_rows,
        _["ncols"] = B.n_cols
    ) ;

}
')
foo( c(1, 2, 3 ) )
# $nrows
# [1] 3
# 
# $ncols
# [1] 1

Three cheers for the first official `sourceCpp()`-using answer :) — Dirk Eddelbuettel, Nov 14 '12 at 16:06
so my option's are adding `A <- matrix(A, ncol = 5);` or using `repmat` and returning a matrix? — pyCthon, Nov 14 '12 at 18:44

score 4 · Answer 2 · answered Nov 14 '12 at 15:53

4

You are trying compound expression involving several heavily templated libraries. That can go wrong. I;d recommended to do it in pieces:

Make sure you have the matrix A you expect passed down to the embedded R
Make sure the function calls worked right, check the result.
Important: check the result type. A matrix should come back fine.
Get the result back to C++.
Get it to Rcpp.
Use RcppArmadillo marshaling to get to Armadillo.

In principle, this should work. The devil is in the detail, as always.

answered Nov 14 '12 at 15:53

Dirk Eddelbuettel

360,940
56
644
725

ah great suggestion's, i've checked the result type with `R.parseEval("str(A)");` and found that it read my matrix as a vector! `num [1:25]` – pyCthon Nov 14 '12 at 16:01
And meanwhile Romain proves my hunch in 3. was right. C++ remains remains a strongly-typed language... – Dirk Eddelbuettel Nov 14 '12 at 16:09

Understanding passing armadillo matrices to R functions via RInside

2 Answers2