Installing Pywr

Pywr should work on Python 3.7 (or later) on Windows, Linux or OS X.

Building Pywr from source requires a working C compiler. It has been built successfully with MSVC on Windows, GCC on Linux and clang/LLVM on OS X.

The easiest way to install Pywr is using the Anaconda Python distribution. See instructions below on Installing binary packages with Anaconda.


Pywr has several external dependencies, listed below.

Installing (in general)

When installing Pywr you must specific which solvers to build. This is done by passing --with-<solver> as an argument to The following command will build and install Pywr with both the GLPK and lpsolve solvers:

python install --with-glpk --with-lpsolve

To install Pywr in-place in developer mode, use the develop command instead of install. This is only useful if you plan to modify the Pywr source code and is not required for general use.

python develop --with-glpk --with-lpsolve

Installing binary wheels with pip

Binary wheel distributions of Pywr are hosted on Pypi for Windows and Linux.

pip install pywr

Installing binary packages with Anaconda

A binary distribution of Pywr is provided for 3.7+ (64-bit) on Windows, Linux and OS X for the Anaconda Python distribution. Note that this release may lag behind the development version.

You will need to install and configure Anaconda before proceeding. The conda 30-minute test drive is a good place to start.

Pywr is hosted on the conda-forge channel.

conda config --add channels conda-forge
conda install pywr

Installing from source with Anaconda

It’s possible to install the dependencies as Anaconda packages, but still build from source. This is only required if you want to keep up with development versions, rather than using the binaries done for releases. In this case you need to specify the include and library paths in your environment as the libraries will be installed in a non-standard location. This can be done by passing the relevant flags to As an example, the following batch script should work on Windows (with a similar approach taken on Linux/macOS)

python build_ext -I"%LIBRARY_INC%" -L"%LIBRARY_LIB%" --inplace --with-glpk --with-lpsolve install

Installing on Windows

To build Pywr from source on Windows you must first install and configure the MSVC compiler. See the instructions on this blog. It is important that you install the correct version of MSVC to correspond with your Python version.

Binaries for GLPK are available from the WinGLPK project. This includes the MSVC solution files if you want to build GLPK yourself on Windows, although the prebuilt binaries are fine.

Binaries for lpsolve are available from the lpsolve sourceforge website.

Installing on Linux

No special instructions required. Follow instructions as for installing (in general) to build from source. A conda package is also available.


The following commands should install the GLPK and lpsolve libraries:

sudo apt-get install libgmp3-dev libglpk-dev glpk
sudo apt-get install liblpsolve55-dev lp-solve

The Ubuntu package for lpsolve includes a static library which can confuse the compiler. The easiest work-around is to remove it:

sudo rm /usr/lib/liblpsolve55.a
sudo ln -s /usr/lib/lp_solve/ /usr/lib/

Installing on OS X

Follow instructions as for installing (in general) to build from source. A conda package is also available.

If external libraries are located in a non-standard location you either need to set the DYLD_LIBRARY_PATH environment variable at runtime:

export DYLD_LIBRARY_PATH=/path/to/library/directory

Alternatively (and recommended) set the rpath of the extension during compilation.

export CFLAGS="-Wl,-rpath,/path/to/library/directory"

You may also need to specify the location of the library headers:

export C_INCLUDE_PATH=/path/to/include/directory

Examples of the above can be seen in the conda recipe (see conda-recipe/

The dependencies (GLPK and/or lpsolve) can be built from source manually, or installed using Homebrew.

Development and testing

The source code for Pywr is managed using Git and is hosted on GitHub: .

There are a collection of unit tests for Pywr written using pytest. These can be run using:

pytest tests

This will run all avaialble tests using the default solver. A specific solver can be tested by specifying the PYWR_SOLVER environment variable:

PYWR_SOLVER=lpsolve pytest tests

Continuous Integration

Pywr is automatically built and tested on Linux and Windows using Travis-CI and AppVeyor (respectively).

Creating a pull request on GitHub will automatically trigger a build.

Both services install Pywr using the Anaconda Python distribution, as this was the easiest way to install all the dependencies.