Python interface for the igraph library

igraph is a library for creating and manipulating graphs. It is intended to be as powerful (ie. fast) as possible to enable the analysis of large graphs.

This repository contains the source code to the Python interface of igraph.

Since version 0.10.2, the documentation is hosted on readthedocs. Earlier versions are documented on our old website.

igraph is a collaborative work of many people from all around the world — see the list of contributors here.

Installation from PyPI

We aim to provide wheels on PyPI for most of the stock Python versions; typically at least the three most recent minor releases from Python 3.x. Therefore, running the following command should work without having to compile anything during installation:

pip install igraph

See details in Installing Python Modules.

Installation from source with pip on Debian / Ubuntu and derivatives

If you need to compile igraph from source for some reason, you need to install some dependencies first:

sudo apt install build-essential python-dev libxml2 libxml2-dev zlib1g-dev

and then run

pip install igraph

This should compile the C core of igraph as well as the Python extension automatically.

Installation from source on Windows

It is now also possible to compile igraph from source under Windows for Python 3.7 and later. Make sure that you have Microsoft Visual Studio 2015 or later installed, and of course Python 3.7 or later. First extract the source to a suitable directory. If you launch the Developer command prompt and navigate to the directory where you extracted the source code, you should be able to build and install igraph using python setup.py install

You may need to set the architecture that you are building on explicitly by setting the environment variable

set IGRAPH_CMAKE_EXTRA_ARGS=-A [arch]

where [arch] is either Win32 for 32-bit builds or x64 for 64-bit builds. Also, when building in MSYS2, you need to set the SETUPTOOLS_USE_DISTUTILS environment variable to stdlib; this is because MSYS2 uses a patched version of distutils that conflicts with setuptools >= 60.0.

Enabling GraphML

By default, GraphML is disabled, because libxml2 is not available on Windows in the standard installation. You can install libxml2 on Windows using vcpkg. After installation of vcpkg you can install libxml2 as follows

vcpkg.exe install libxml2:x64-windows-static-md

for 64-bit version (for 32-bit versions you can use the x86-windows-static-md triplet). You need to integrate vcpkg in the build environment using

vcpkg.exe integrate install

This mentions that

CMake projects should use: -DCMAKE_TOOLCHAIN_FILE=[vcpkg build script]

which we will do next. In order to build igraph correctly, you also need to set some other environment variables before building igraph:

set IGRAPH_CMAKE_EXTRA_ARGS=-DVCPKG_TARGET_TRIPLET=x64-windows-static-md -DCMAKE_TOOLCHAIN_FILE=[vcpkg build script]
set IGRAPH_EXTRA_LIBRARY_PATH=[vcpkg directory]/installed/x64-windows-static-md/lib/
set IGRAPH_STATIC_EXTENSION=True
set IGRAPH_EXTRA_LIBRARIES=libxml2,lzma,zlib,iconv,charset
set IGRAPH_EXTRA_DYNAMIC_LIBRARIES: wsock32,ws2_32

You can now build and install igraph again by simply running python setup.py build. Please make sure to use a clean source tree, if you built previously without GraphML, it will not update the build.

Linking to an existing igraph installation

The source code of the Python package includes the source code of the matching igraph version that the Python interface should compile against. However, if you want to link the Python interface to a custom installation of the C core that has already been compiled and installed on your system, you can ask setup.py to use the pre-compiled version. This option requires that your custom installation of igraph is discoverable with pkg-config. First, check whether pkg-config can tell you the required compiler and linker flags for igraph:

pkg-config --cflags --libs igraph

If pkg-config responds with a set of compiler and linker flags and not an error message, you are probably okay. You can then proceed with the installation using pip after setting the environment variable named IGRAPH_USE_PKG_CONFIG to 1 to indicate that you want to use an igraph instance discoverable with pkg-config:

IGRAPH_USE_PKG_CONFIG=1 pip install igraph

Alternatively, if you have already downloaded and extracted the source code of igraph, you can run setup.py directly:

IGRAPH_USE_PKG_CONFIG=1 python setup.py build
IGRAPH_USE_PKG_CONFIG=1 python setup.py install

(Note that you need the IGRAPH_USE_PKG_CONFIG=1 environment variable for both invocations, otherwise the call to setup.py install would still build the vendored C core instead of linking to an existing installation).

This option is primarily intended for package maintainers in Linux distributions so they can ensure that the packaged Python interface links to the packaged igraph library instead of bringing its own copy.

It is also useful on macOS if you want to link to the igraph library installed from Homebrew.

Due to the lack of support of pkg-config on Windows, it is currently not possible to build against an external library on Windows.

Warning: the Python interface is guaranteed to work only with the same version of the C core that is vendored inside the vendor/source/igraph folder. While we try hard not to break API or ABI in the C core of igraph between minor versions in the 0.x branch and we will keep on doing so for major versions once 1.0 is released, there are certain functions in the C API that are marked as experimental (see the documentation of the C core for details), and we reserve the right to break the APIs of those functions, even if they are already exposed in a higher-level interface. This is because the easiest way to test these functions in real-life research scenarios is to expose them in one of the higher level interfaces. Therefore, if you unbundle the vendored source code of igraph and link to an external version instead, we can make no guarantees about stability unless you link to the exact same version as the one we have vendored in this source tree.

If you are curious about which version of the Python interface is compatible with which version of the C core, you can look up the corresponding tag in Github and check which revision of the C core the repository points to in the vendor/source/igraph submodule.

Compiling the development version

If you want to install the development version, the easiest way to do so is to install it using

pip install git+https://github.com/igraph/python-igraph

This automatically fetches the development version from the repository, builds the package and installs it. By default, this will install the Python interface from the main branch, which is used as the basis for the development of the current release series. Unstable and breaking changes are being made in the develop branch. You can install this similarly by doing

pip install git+https://github.com/igraph/python-igraph@develop

In addition to git, the installation of the development version requires some additional dependencies, read further below for details.

For more information about installing directly from git using pip see https://pip.pypa.io/en/stable/topics/vcs-support/#git.

Alternatively, you can clone this repository locally. This repository contains a matching version of the C core of igraph as a git submodule. In order to install the development version from source, you need to instruct git to check out the submodules first:

git submodule update --init

Compiling the development version additionally requires flex and bison. You can install those on Ubuntu using

sudo apt install bison flex

On macOS you can install these from Homebrew or MacPorts. On Windows you can install winflexbison3 from Chocolatey.

Then, running the setup script should work if you have a C compiler and the necessary build dependencies (see also the previous section):

python setup.py build

You can install it using

python setup.py install

Running unit tests

Unit tests can be executed from within the repository directory with tox or with the built-in unittest module:

python -m unittest

Contributing

Contributions to igraph are welcome!

If you want to add a feature, fix a bug, or suggest an improvement, open an issue on this repository and we'll try to answer. If you have a piece of code that you would like to see included in the main tree, open a PR on this repo.

To start developing igraph, follow the steps above about installing the development version. Make sure that you do so by cloning the repository locally so that you are able to make changes.

For easier development, you can install igraph in development mode so your changes in the Python source code are picked up automatically by Python:

python setup.py develop

Changes that you make to the Python code do not need any extra action. However, if you adjust the source code of the C extension, you need to rebuild it by running python setup.py develop again. Compilation of the C core of igraph is cached in vendor/build and vendor/install so subsequent builds are much faster than the first one as the C core does not need to be recompiled.

Notes

Supported Python versions

We aim to keep up with the development cycle of Python and support all official Python versions that have not reached their end of life yet. Currently this means that we support Python 3.7 to 3.11, inclusive. Please refer to this page for the status of Python branches and let us know if you encounter problems with igraph on any of the non-EOL Python versions.

Continuous integration tests are regularly executed on all non-EOL Python branches.

PyPy

This version of igraph is compatible with PyPy and is regularly tested on PyPy with tox. However, the PyPy version falls behind the CPython version in terms of performance; for instance, running all the tests takes ~5 seconds on my machine with CPython and ~15 seconds with PyPy. This can probably be attributed to the need for emulating CPython reference counting, and does not seem to be alleviated by the JIT.

There are also some subtle differences between the CPython and PyPy versions:

• Docstrings defined in the C source code are not visible from PyPy.

• GraphBase is hashable and iterable in PyPy but not in CPython. Since GraphBase is internal anyway, this is likely to stay this way.