Installation instructions

Installing a pre-built package

Installing via conda

The conda-forge channel provides pre-built CVXOPT packages for Linux, macOS, and Windows that can be installed using conda:

conda install -c conda-forge cvxopt

These pre-built packages are linked against OpenBLAS and include all the optional extensions (DSDP, FFTW, GLPK, and GSL).

Installing via pip

A pre-built binary wheel package can be installed using pip:

pip install cvxopt

Wheels for Linux:

  • are available for Python 2.7, 3.4, 3.5, 3.6, and 3.7 (32 and 64 bit)
  • are linked against OpenBLAS
  • include all optional extensions (DSDP, GLPK, GSL, and FFTW)

Wheels for macOS:

  • are available for Python 2.7, 3.4, 3.5, 3.6, and 3.7 (universal binaries)
  • are linked against BLAS/LAPACK from the Accelerate framework
  • include all optional extensions (DSDP, GLPK, GSL, and FFTW)

Wheels for Windows:

  • are available for Python 27, 3.4, 3.5, 3.6, and 3.7 (64 bit only)
  • are linked against MKL
  • Python 3.5+ wheels include the optional extension GLPK (wheels for Python 2.7 and 3.4 include none of the optional extensions)

Building and installing from source

Required and optional software

The package requires version 2.7 or 3.x of Python, and building from source requires core binaries and header files and libraries for Python.

The installation requires BLAS and LAPACK. Using an architecture optimized implementation such as ATLAS, OpenBLAS, or MKL is recommended and gives a large performance improvement over reference implementations of the BLAS and LAPACK libraries.

The installation also requires SuiteSparse. We recommend linking against a shared SuiteSparse library. It is also possible to build the required components of SuiteSparse when building CVXOPT, but this requires the SuiteSparse source which is no longer included with CVXOPT and must be downloaded separately.

The following software libraries are optional.

Installation

CVXOPT can be installed globally (for all users on a UNIX/Linux system) using the command:

python setup.py install

It can also be installed locally (for a single user) using the command:

python setup.py install --user

To test that the installation was successful, run the included tests using:

python -m unittest discover -s tests

or alternatively, if nose is installed:

nosetests

If Python does not issue an error message, the installation was successful.

It is also possible to install CVXOPT from source using pip:

pip install cvxopt --no-binary cvxopt

Additional information can be found in the Python documentation.

Customizing the setup script

If needed, the default compilation can be customized by editing setup.py or by means of environment variables. The following variables in the setup script can be modified:

  • BLAS_LIB_DIR: the directory containing the LAPACK and BLAS libraries.
  • BUILD_GSL: set this variable to 1 if you would like to use the GSL random number generators for constructing random matrices in CVXOPT. If BULD_GSL is 0, the Python random number generators will be used instead.
  • GSL_LIB_DIR: the directory containing libgsl.
  • GSL_INC_DIR: the directory containing the GSL header files.
  • BUILD_FFTW: set this variable to 1 to install the cvxopt.fftw module, which is an interface to FFTW.
  • FFTW_LIB_DIR: the directory containing libfftw3.
  • FFTW_INC_DIR: the directory containing fftw.h.
  • BUILD_GLPK: set this variable to 1 to enable support for the linear programming solver GLPK.
  • GLPK_LIB_DIR: the directory containing libglpk.
  • GLPK_INC_DIR: the directory containing glpk.h.
  • BUILD_DSDP: set this variable to 1 to enable support for the semidefinite programming solver DSDP.
  • DSDP_LIB_DIR: the directory containing libdsdp.
  • DSDP_INC_DIR: the directory containing dsdp5.h.
  • SUITESPARSE_LIB_DIR: the directory containing SuiteSparse libraries.
  • SUITESPARSE_INC_DIR: the directory containing SuiteSparse header files.
  • SUITESPARSE_SRC_DIR: the directory containing SuiteSparse source. The variables SUITESPARSE_LIB_DIR and SUITESPARSE_INC_DIR are ignored and relevant parts of SuiteSparse are build from source when SUITESPARSE_SRC_DIR is specified.
  • MSVC: set this variable to 1 if compiling with MSVC 14 or later

Each of the variables can be overridden by specifying an environment variable with the prefix CVXOPT_. For example, the following command installs CVXOPT locally with BUILD_FFTW=1:

CVXOPT_BUILD_FFTW=1 python setup.py install --user

This approach also works with pip:

export CVXOPT_BUILD_FFTW=1
pip install cvxopt --no-binary cvxopt

Support for the linear, second-order cone, and quadratic programming solvers in MOSEK is automatically enabled if both MOSEK and its Python interface are installed.

Ubuntu/Debian

Building CVXOPT from source in Debian/Ubuntu requires the packages build-essential and python-dev as well as BLAS and LAPACK library packages such as

  • libopenblas-dev
  • libatlas-dev
  • libblas-dev and liblapack-dev

If multiple BLAS and LAPACK libraries have been installed, you can verify the current configuration using the following commands:

update-alternatives --config libblas.so.3
update-alternatives --config liblapack.so.3

As of Ubuntu 16.04, SuiteSparse can be installed as a dynamic library by installing the libsuitesparse-dev package. Alternatively, if SuiteSparse is not available as a dynamic library, the SuiteSparse source must be available.

To build the optional CVXOPT extensions (DSDP, FFTW, GLPK, and GSL), the following packages should be installed as well:

  • libdsdp-dev
  • libfftw3-dev
  • libglpk-dev
  • libgsl-dev

When all the necessary Ubuntu packages have been installed, CVXOPT can be built with all extensions in Ubuntu 16.04 (or later) as follows:

git clone https://github.com/cvxopt/cvxopt.git
cd cvxopt
git checkout `git describe --abbrev=0 --tags`
export CVXOPT_BUILD_DSDP=1    # optional
export CVXOPT_BUILD_FFTW=1    # optional
export CVXOPT_BUILD_GLPK=1    # optional
export CVXOPT_BUILD_GSL=1     # optional
python setup.py install

To use the Intel MKL library instead of ATLAS or OpenBLAS, include the following commands before running python setup.py install:

pip install mkl
MKLLIB=mkl_rt
PYDIR=`pip show mkl | grep Location | cut -d' ' -f 2`
MKLDIR=`grep lib${MKLLIB} $PYDIR/mkl*/RECORD | cut -d, -f1`
PREFIX_LIB=`dirname $PYDIR/$MKLDIR`
export CVXOPT_LAPACK_LIB=${MKLLIB}
export CVXOPT_BLAS_LIB=${MKLLIB}
export CVXOPT_BLAS_LIB_DIR=${PREFIX_LIB}
export CVXOPT_BLAS_EXTRA_LINK_ARGS="-L${PREFIX_LIB};-Wl,-rpath,${PREFIX_LIB};-l${MKLLIB}"

In older versions of Ubuntu where SuiteSparse is not available as a dynamic library, the necessary SuiteSparse components can be built with CVXOPT by downloading the SuiteSparse source and setting CVXOPT_SUITESPARSE_SRC_DIR to the SuiteSparse source directory:

wget http://faculty.cse.tamu.edu/davis/SuiteSparse/SuiteSparse-5.2.0.tar.gz
tar -xf SuiteSparse-5.2.0.tar.gz
export CVXOPT_SUITESPARSE_SRC_DIR=$(pwd)/SuiteSparse
git clone https://github.com/cvxopt/cvxopt.git
cd cvxopt
git checkout `git describe --abbrev=0 --tags`
export CVXOPT_BUILD_DSDP=1    # optional
export CVXOPT_BUILD_FFTW=1    # optional
export CVXOPT_BUILD_GLPK=1    # optional
export CVXOPT_BUILD_GSL=1     # optional
python setup.py install

macOS

Building CVXOPT from source in macOS requires the Command-line tools which can be installed using the command:

xcode-select -p

With Homebrew

Homebrew users can build CVXOPT with FFTW, GLPK, and GSL as follows:

brew install gsl fftw suite-sparse glpk
git clone https://github.com/cvxopt/cvxopt.git
cd cvxopt
git checkout `git describe --abbrev=0 --tags`
export CVXOPT_BUILD_FFTW=1    # optional
export CVXOPT_BUILD_GLPK=1    # optional
export CVXOPT_BUILD_GSL=1     # optional
python setup.py install

To use OpenBLAS instead of the built-in BLAS/LAPACK libraries, include the following commands before running python setup.py install:

brew install openblas
export CVXOPT_BLAS_LIB_DIR=/usr/local/opt/openblas/lib
export CVXOPT_BLAS_LIB=openblas
export CVXOPT_LAPACK_LIB=openblas

Alternatively, to use the Intel MKL library, include the following commands before running python setup.py install:

pip install mkl
MKLLIB=mkl_rt
PYDIR=`pip show mkl | grep Location | cut -d' ' -f 2`
MKLDIR=`grep lib${MKLLIB} $PYDIR/mkl*/RECORD | cut -d, -f1`
PREFIX_LIB=`dirname $PYDIR/$MKLDIR`
if [[ $OSTYPE == darwin* ]]; then
    install_name_tool -change @rpath/libiomp5.dylib @loader_path/libiomp5.dylib ${PREFIX_LIB}/libmkl_intel_thread.dylib
fi
export CVXOPT_LAPACK_LIB=${MKLLIB}
export CVXOPT_BLAS_LIB=${MKLLIB}
export CVXOPT_BLAS_LIB_DIR=${PREFIX_LIB}
export CVXOPT_BLAS_EXTRA_LINK_ARGS="-L${PREFIX_LIB};-Wl,-rpath,${PREFIX_LIB};-l${MKLLIB}"
pip install git+https://github.com/cvxopt/cvxopt

Without Homebrew

If SuiteSparse is not available as a dynamic library, the necessary SuiteSparse components can be built with CVXOPT by downloading the SuiteSparse source and setting CVXOPT_SUITESPARSE_SRC_DIR to the SuiteSparse source directory:

wget http://faculty.cse.tamu.edu/davis/SuiteSparse/SuiteSparse-5.2.0.tar.gz
tar -xf SuiteSparse-5.2.0.tar.gz
export CVXOPT_SUITESPARSE_SRC_DIR=$(pwd)/SuiteSparse
git clone https://github.com/cvxopt/cvxopt.git
cd cvxopt
git checkout `git describe --abbrev=0 --tags`
python setup.py install

Windows

We will assume that Python (64 bit), git, wget, and 7-zip are installed and in the search path. These can be installed with the Chocolatey package manager:

choco install -y wget git python2 7zip.commandline

We also will assume that the environment variable %PYTHON% contains the path to the Python installation directory, e.g.,

set PYTHON=c:\PythonXX

or alternatively,

for /f %i in ('python -c "import sys, os; print(os.path.dirname(sys.executable))"') do set PYTHON=%i

where % must be replaced by %% if the above line is included in a batch file.

Finally, we will assume that pip is available; if it is not, it can now be installed with easy_install:

%PYTHON%\Scripts\easy_install pip

Python 2.7, Python 3.4

CVXOPT can be built for Windows (64 bit) with the Mingwpy toolchain and MKL. Note that Mingwpy currently only supports Python version 2.7 through 3.4.

Open the Command Prompt and execute the following commands:

rem Download SuiteSparse source
wget http://faculty.cse.tamu.edu/davis/SuiteSparse/SuiteSparse-5.2.0.tar.gz
7z x SuiteSparse-5.2.0.tar.gz
7z x SuiteSparse-5.2.0.tar
set CVXOPT_SUITESPARSE_SRC_DIR=%CD%\SuiteSparse

rem Install MKL
pip install mkl
set CVXOPT_BLAS_LIB_DIR=%PYTHON%\Library\lib
set CVXOPT_BLAS_LIB=mkl_rt
set CVXOPT_LAPACK_LIB=mkl_rt

rem Install mingwpy Python extension using pip
pip install -i https://pypi.anaconda.org/carlkl/simple mingwpy

rem Clone CVXOPT repository, compile, install, and run tests
git clone https://github.com/cvxopt/cvxopt.git
cd cvxopt
for /f %%a in ('git describe --abbrev^=0 --tags') do git checkout %%a
python setup.py build --compiler=mingw32
python setup.py install
python -m unittest discover -s tests

Python 3.5+

CVXOPT 1.2.0+ can be built for Windows (64 bit) with MSVC14 and MKL.

Open the Command Prompt and execute the following commands:

rem Download SuiteSparse source
wget http://faculty.cse.tamu.edu/davis/SuiteSparse/SuiteSparse-5.2.0.tar.gz
7z x SuiteSparse-5.2.0.tar.gz
7z x SuiteSparse-5.2.0.tar
set CVXOPT_SUITESPARSE_SRC_DIR=%CD%\SuiteSparse

rem Install MKL
pip install mkl
set CVXOPT_BLAS_LIB_DIR=%PYTHON%\Library\lib
set CVXOPT_BLAS_LIB=mkl_rt
set CVXOPT_LAPACK_LIB=mkl_rt

rem Clone CVXOPT repository, compile, install, and run tests
git clone https://github.com/cvxopt/cvxopt.git
cd cvxopt
for /f %%a in ('git describe --abbrev^=0 --tags') do git checkout %%a
set CVXOPT_MSVC=1
python setup.py build --compiler=msvc
python setup.py install
python -m unittest discover -s tests