SuiteSparse Installation: Cannot open include file: 'regex.h'

I am having issues installing SuiteSparse in Matlab. I am using the 5.3 version, downloaded from http://faculty.cse.tamu.edu/davis/suitesparse.html, and using the Visual Studio 2017 C++ compiler. When I run SuiteSparse_install, I get the following error:

c:\users\dylan\documents\school\masters\suitesparse\metis-5.1.0\gklib\GKlib.h(68): fatal error C1083: Cannot open include file: 'regex.h': No such file or directory

And UMFPACK, KLU, CHOLMOD, SuiteSparseQR fail to install. Has anyone got any experience with this, or could point me in the right direction for assistance? I need the sparse2 and spqr functions for my code, as I have a very large over-determined, sparse system to solve, and it is giving Matlab's mldivide a rough time. Thanks!

[Jorge Barrasa Fano]

Hi Dylan,

I've the exact same problem. Have you managed to solve it so far?

Thanks,

Jorge

[Dylan Zaluski]

Hi Jorge,

Yes, I actually did get this resolved some time ago, after talking to Tim Davis (SuiteSparse's creator). The issue lies with the METIS package that is included with SuiteSparse. When you download a copy, before building the package, delete the entire folder named "METIS" (might be a version number along with the name), and then try building it again. That worked like a charm for me.

[Jorge Barrasa Fano]

Hi Dylan,

Thanks for your reply :) I ended up doing the same and it worked.

[Humberto Brambila de Salles]

Dylan, you are the real MVP! I've been trying this for so long! Thank you so much!!

Answer 1

[Royi Avital]

Why would you need SuiteSparse in MATLAB?

I though MATLAB uses SuiteSparse to begin with.

[Petros Bogiatzis]

SuiteSparce has a lot (really a lot) more functionalities that are note included in the embended Matlab versions of the functions.

[Royi Avital]

Could you elaborate on that? Such as?

[Dylan Zaluski]

The SuiteSparse package has several additional features and performance enhancements over the versions implemented in the current MATLAB distribution I believe. The sparse2 function is faster for sparse matrix allocation, and more memory efficient. It includes sparse linear solvers that have been successful for me when MATLAB's included solvers (backslash) have failed. I also use its csltsolve and cslsolve functions for quickly solving lower and upper triangular systems in preconditioned conjugate gradient codes.

[Royi Avital]

Interesting. Does it have an equivalent of \ which is faster? Does it have iterative solver as well? Does it handle Mat Vec / Mat Mat multiplication as well?

[Dylan Zaluski]

This really depends on your situation, and the structure of your matrix. It has faster cholesky decomposition functions, as well as a faster QR method. For my own work, I needed the spqr_solve function for solving least-squares problems, and the performance benefit was substantial. You'll just need to try it out for yourself and see if it provides a speedup. It is very easy to install and use. Check the documentation out to find out if it has what you need. And no, it does not provide any improved Mat Vec or Mat Mat product functions. Check out the Intel MKL BLAS library for that.

[Royi Avital]

Dylan, MATLAB uses SuiteSparse for QR of Sprase Matrices to begin with (It's built in). It also uses MKL for BLAS things. Could you show a sample code you get improvement from?