1

How can one extract column labels from an Rcpp::Eigen matrix?

For example, here is some R matrix

mat           <- matrix(1:4,ncol=2)
colnames(mat) <- c("col1","col2")

With Rcpp::NumericMatrix, one can simply call colnames like so:

void fun_NM(const Rcpp::NumericMatrixXd& Mat)
{
  Rcpp::CharacterVector coln = colnames(Mat);
  Rcpp::Rcout << coln << "\n\n";  
}

Then fun_NM(mat) prints "col1" "col2" as it should. Is there a way to access these colnames when the matrix is passes as an Eigen::MatrixXd?

void fun_EM(const Eigen::MatrixXd& Mat)
{
  ?
}

I don't know how RcppEigen converts the R matrix passed to into the Rcpp code to a Eigen::MatrixXd. Does it first converts it to a Rcpp::NumericMatrix? In that case, one could use Rcpp::NumericMatrix as input, extract the columns, and then transform it to Eigen::MatrixXd manually in the code to use the function in the Eigen library.

coatless
  • 20,011
  • 13
  • 69
  • 84
Martin Schab
  • 13
  • 2
  • Column names are additional attributes at a the R level that don't really have meaning at the "linear algebra level" that Eigen (or Armadillo) operate on. They just use indices. I recommend writing a local "pretty-printer" routine that adds column names if you want them. – Dirk Eddelbuettel Dec 11 '20 at 12:56
  • Ok, thank you! The thing is that I would like to use the column names used by the R user in that "pretty printer" function. Is there a way to do that with Eigen::Matrix or must I change the input to Rcpp::NumericMatrix and then convert to Eigen::MatrixXd after reading off the column labels? – Martin Schab Dec 11 '20 at 13:04

1 Answers1

2

It helps to step back. What are names? An attribute. So in R we'd do

mat <- matrix(1:4,ncol=2)
colnames(mat) <- c("col1","col2")
attributes(mat)

Well, turns out in C++ with Rcpp it is just about the same---see several articles at the Rcpp Gallery or answers here.

But, and there is always a but, going to an Eigen matrix drops attributes as the Eigen C++ classes have no such notion. But if you wanted to, you start with a SEXP (or a NumericMatrix), access the attributes and then proceed. A really simple example is below.

Code

#include <RcppEigen.h>

// [[Rcpp::depends(RcppEigen)]]

// [[Rcpp::export]]
void showNames(SEXP m) {
  Rcpp::NumericMatrix nm(m);     // to easily access attributes
  Rcpp::List dmnm = nm.attr("dimnames");
  print(dmnm);

  // continue as before
  Eigen::MatrixXd Mat = Rcpp::as<Eigen::MatrixXd>(m); // make Mat from m as before
  // ... some stuff with Mat
}

/*** R
mat <- matrix(1:4,ncol=2)
colnames(mat) <- c("col1","col2")
attributes(mat)

showNames(mat)
*/

Output

> Rcpp::sourceCpp("~/git/stackoverflow/65251442/answer.cpp")

> 
mat <- matrix(1:4,ncol=2)

> 
colnames(mat) <- c("col1","col2")

> 
attributes(mat)
$dim
[1] 2 2

$dimnames
$dimnames[[1]]
NULL

$dimnames[[2]]
[1] "col1" "col2"



> 
showNames(mat)
[[1]]
NULL

[[2]]
[1] "col1" "col2"

>

You could access just column names the same way.

Dirk Eddelbuettel
  • 360,940
  • 56
  • 644
  • 725