1

I have the following Rcpp code

#include <RcppArmadillo.h>
#define _USE_MATH_DEFINES
#include<cmath>
#include <boost/math/special_functions/bessel.hpp> 

#include <mlpack.h>
#include <mlpack/methods/kmeans.hpp>
// [[Rcpp::depends(RcppEnsmallen)]]
// [[Rcpp::depends(mlpack)]]

// [[Rcpp::plugins(cpp17)]]
// [[Rcpp::depends(RcppArmadillo)]]
// [[Rcpp::plugins(openmp)]]
// [[Rcpp::depends(BH)]]

using namespace arma;
using namespace Rcpp;

#define ARMA_USE_ARPACK 1
#define ARMA_USE_SUPERLU 1
#define tolr 1e-5
#define crossprod(x) symmatu((x).t() * (x))
#define tcrossprod(x) symmatu((x) * (x).t())
#define pi_sq 9.869604401089358


// [[Rcpp::export]]
Rcpp::List NNGP_fit(const int n){
  sp_mat A = sprandu<sp_mat>(n, n, 0.1);
  sp_mat B = A.t()*A;
  
  vec eigval;
  mat eigvec;
  
  eigs_sym(eigval, eigvec, B, n-1);  // find n-1 eigenvalues/eigenvectors
  Rcpp::Rcout<<"flag eig"<<endl;
  
  superlu_opts opts;
  opts.symmetric  = true;
  opts.equilibrate =true;
  vec mu_f,
   y(n, fill::randn);
  
  bool status = spsolve(mu_f,B, y, "superlu",opts);
  Rcpp::Rcout<<"status= "<<status<<endl;
  
  return Rcpp::List::create(Rcpp::Named("covmat") =wrap(B),
                            Rcpp::Named("eigval") =wrap(eigval),
                            Rcpp::Named("eigvec") =wrap(eigvec),
                            Rcpp::Named("y") =wrap(y),
                            Rcpp::Named("mu_f") =wrap(mu_f)
  );
}

I am getting the following output

flag eig
Error: spsolve(): use of SuperLU must be enabled

That is, the code stops working at the spsolve command but executes eigs_sym.

However, removing mlpack by deleting the following solves the problem.

#include <mlpack.h>
#include <mlpack/methods/kmeans.hpp>
// [[Rcpp::depends(RcppEnsmallen)]]
// [[Rcpp::depends(mlpack)]]

So I believe I have linked superlu correctly.

This is only a reproducible part of the code and I want to use mlpack in my project. How can I solve this issue? I also need functions from the boost library for the project and kept that header as well.

Improve this question

1 Answer 1

Reset to default
1

When you work with SuperLU you are leaving the (very convenient) world of header-only libraries such as the ones you use here:

  • Armadillo via RcppArmadillo (as it needs no linking to LAPACK/BLAS when used from R)
  • ensmallen via RcppEnsmall as it is header-only
  • Boost via BH as it concentrates on a header-only subset.
  • mlpack which since version 4.0 works header-only with R.

But sparse linear algebra needs linking. This is explained and demonstrated over by an Rcpp Gallery article from two of the RcppArmadillo authors.

In general it is often advisable to solve smaller subset problems first. You have not real mlpack issue here: you have an issue using RcppArmadillo with sparse decomposition and for that you to link with -lsuperlu as the Rcpp Gallery describes and shows. I would try to get that to work, and then try to proceed to mlpack issues.

PS 1 Your code has another issue preventing it from running: the behavior-altering #define must come before the #include so please move

#define ARMA_USE_ARPACK 1
#define ARMA_USE_SUPERLU 1

to the very top. It then works as the run-time has a normal coding issue:

> Rcpp::sourceCpp("soquestion20241227.cpp")
> set.seed(123)
> NNGP_fit(3)
flag eig
status= 0
Error in getClass("dgCMatrix") : “dgCMatrix” is not a defined class

Note that in order to get this far I had to enable linking with SuperLU via the envoirnment variable, and I validated against the Rcpp Gallery example code:

> Sys.setenv("PKG_LIBS"="-lsuperlu")
> Rcpp::sourceCpp("gallery20170217.cpp")
> superLuDemo()
Done.
> 
>

It is important to set this before the first time you call Rcpp::sourceCpp() as some of these things appear to be cached per session.

PS 2 Here is a reproducible success. Note that I load the Matrix package (for sparse matrix support) and set the seed to be reproducible.

> library(Matrix)
> Sys.setenv("PKG_LIBS"="-lsuperlu")
> Rcpp::sourceCpp("soquestion20241227.cpp")
> set.seed(123)
> NNGP_fit(5)
flag eig
status= 0
$covmat
5 x 5 sparse Matrix of class "dgCMatrix"
                             
[1,] 0.0827008 . . . 0.226699
[2,] .         . . . .       
[3,] .         . . . .       
[4,] .         . . . .       
[5,] 0.2266988 . . . 0.788687

$eigval
            [,1]
[1,] 7.05517e-33
[2,] 6.13069e-17
[3,] 1.61746e-02
[4,] 8.55213e-01

$eigvec
             [,1]         [,2]         [,3]         [,4]
[1,]  1.97882e-16 -7.23362e-15  9.59537e-01 -2.81582e-01
[2,]  1.64671e-01  7.22721e-01  5.40204e-15  2.43186e-17
[3,] -3.27270e-02 -6.76148e-01 -5.07065e-15 -1.23420e-17
[4,] -9.85805e-01  1.43172e-01  1.05140e-15  1.94107e-18
[5,]  3.29803e-17  2.09006e-15 -2.81582e-01 -9.59537e-01

$y
           [,1]
[1,]  0.0699539
[2,]  0.9767215
[3,]  2.1650415
[4,] -1.8262054
[5,]  0.5804146

$mu_f
     [,1]

>
Improve this answer
Sign up to request clarification or add additional context in comments.

9 Comments

I have already seen the gallery you mentioned. I am able to resolve the issue without using mlpack. If I simply delete the following, it works just fine ``` #include <mlpack.h> #include <mlpack/methods/kmeans.hpp> // [[Rcpp::depends(RcppEnsmallen)]] // [[Rcpp::depends(mlpack)]] ``` But I need mlpack also, so need help on using both.
I would still recommend to minimize your question further. If the gallery piece works for you, great -- build up from there. I mentioned it because you did not show the linking instructions in your question and this cannot work without linking as SuperLU is not header-only. Further, your error message says just that: no SuperLU code (as you have not linked to it). You simply confuse header-only and not-header-only-but-linking but sadly wishing alone does not make the linking requirement go away. Good luck.
I am not able to build up from the example :). The example works for me so I believe I am able to link superlu. Problem arises when I include the mlpack headers.
You can repeat the same statements as many times as you want, they still miss the fact that you did not do this to enable linking: Sys.setenv("PKG_LIBS"="-lsuperlu") (which works if libsuperlu.* is in a system library directory, else you need -L....location here...). SuperLU needs linking. Just because something compiles does not guarantee it is executable if the (required) object code is not linked. Header-only is great, not all libraries are header-only. That's your issue here: SuperLU needs linking.
See my expanded answer. You had other issues too. But it's compiling (and linking) and running (and then running into a coding issues with sparse matrices I will leave for you).
|

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.