Cholmod and Scikit-Sparse for Windows

Cholmod is a powerful package for sparse matrix calculation and Scikit-Sparse is a python interface. It is convenient to set up the Cholmod and Scikit-Sparse (CSC) environment in Linux and Mac OS, but it may be troublesome in Windows. Thanks to jlblancoc's package suitesparse-metis-for-windows, I make a modification for the python interface. Here are the instructions.

1. Overview

It takes two steps to set up the CSC environment.

1. Compile the *.h, *.lib and *.dll files of Cholmod. It can be included and linked for C/C++.
2. Compile the python package Scikit-Sparse.

This repository has been tested on

• Windows 10 64bit
• Anaconda 3
• Microsoft Visual C++ 17

This repository is not applicable to Python 2 (Anaconda 2) on Windows 10. Because in the Step II, compiling C codes for Python 2 requires VC 9.0 (VS 2008), which can not be installed on Windows 10. The compiler used in the Step I is supposed to be the same as the one used in the Step II.

Because metis package has many errors when compiling, I removed the metis package and modified the CSC_ROOT/CMakeLists.txt.

2. Instructions

2.1 Preparation

1. Microsoft Visual Studio 2017 including VC 14 (VS 2015 may be Ok). C++ compiler tools are required. You can add them by the Stack Overflow post.
2. Anaconda 3.6 64bit including numpy and scipy.
3. Download CMake.
4. Clone or Download this repository, say CSC_ROOT. If you require the latest version, just download them and merge them with the target folders.
   • suitesparse-metis-for-windows version: 1.3.1
   • scikit-sparse version: 0.4.4. If you replace it by a new package, don't forget to modify the setup.py to link the headers and libraries.
   • SuiteSparse version: 5.3.0. It is under the suitesparse-metis-for-windows-1.3.1/

2.1 Compile Cholmod

1. Run CMake, then:
   1. Set the "Source code" folder to CSC_ROOT/suitesparse-metis-for-windows-1.3.1.
   2. Set the "Build the binaries" folder to CSC_ROOT/suitesparse-metis-for-windows-1.3.1/build.
   3. Click "Configure" and select "Visual Studio 15 2017 Win64" (for 64bit OS). Tips: Make sure to build in x64 mode. "HAVE_COMPLEX" is not recommended.
   4. Click "Generate".
   5. Click "Open Project" to open the "SuiteSparseProject" solution.
2. In Visual Studio, build the "INSTALL" project in "Release" mode. It may show hundreds of warnings, but it's OK. (Note: according to this issue#2 report, select "Release" mode is necessary in case of missing cholmod.lib.)
3. CSC_ROOT/suitesparse-metis-for-windows-1.3.1/build/install contains the generated libraries.
   • *.h header files are in CSC_ROOT/suitesparse-metis-for-windows-1.3.1/build/install/include/suitesparse
   • *.lib libraries are in CSC_ROOT/suitesparse-metis-for-windows-1.3.1/build/install/lib64 and CSC_ROOT/suitesparse-metis-for-windows-1.3.1/build/install/lib64/lapack_blas_windows
   • *.dll dynamic libraries are in CSC_ROOT/suitesparse-metis-for-windows-1.3.1/build/install/lib64/lapack_blas_windows.
4. Now, the Cholmod package has been set up for C/C++.

2.2 Compile Scikit-Sparse

1. Open "PowerShell" and type cd scikit-sparse-0.4.4.
2. Type python setup.py build to compile the package. The setup.py has been modified to link the headers and libraries.
3. Type python setup.py install to copy the executable code into the Anaconda folders.

2.3 Testing

Congratulations! Everything is done! In python command. Type

from sksparse.cholmod import cholesky

If nothing happens, the installation is done! For detailed usage and test, please refer to scikit-sparse-0.4.4/sksparse/test_cholmod.py.

3 Update Logs

Give me a star if you like this repository. Feel free to open an issue or start an pull request.

• Thanks to the Issue 4, the scikit-sparse is upgraded from 0.4.3 to 0.4.4. Cython is not included in scikit-sparse 0.4.4.
• Cython is added by Pull 3.