Installation notes for graph-tool


Intent of this post is to document early efforts to get peartree, my library for generated directed network graphs based on public transit schedule data into graph-tool. What is graph-tool? It’s a high-performance C++ library in which core graph algorithms are implemented with an eye towards speed. As a result, computation-intensive network graph calculations can be executed quickly. The library also features Python bindings, enabling a more convenient interface between the existing peartree graph network product and graph-tool’s C++-based model.

This post is intended to be a precursor to a post I am writing on how to convert a NetworkX graph into a graph-tool graph.

Why a blog post?

Setting up graph-tool can be difficult! While the author provides Docker images that enable a user to get up and running with, say, a notebook fairly quickly I wanted to document how I got the tool set up within the development environment I use for peartree.

Installing graph-tool

At the time of this post’s writing, the current Dockerfile configuration for the peartree development environment looks like this. The latest image is available on Docker Hub, here.

The TL;DR is that I’m running Python 3.6 on Debian. I believe the image the author publishes is on Alpine. Regardless, I did not find I could easily/satisfactorily port the peartree development environment. As a result, I wanted to get my development environment as close as I could to conforming with a distribution of the graph-tool C library as possible.

To do so, I made sure I was using Debian Stretch. This allowed me to use the Stretch distribution of the library that the author publishes.

With that, I was able to update my sources.list file with that distribution’s address:

echo "deb stretch main" | tee -a /etc/apt/sources.list && \
echo "deb-src stretch main" | tee -a /etc/apt/sources.list

At this point, the author, in the installation instructions, informs the user to verify the packages with the following public key:

apt-key adv --keyserver --recv-key 612DEFB798507F25

During the installation process, I was prompted with a different public key, this one associated with an Ubuntu keyserver (shown below).

apt-key adv --keyserver --recv-keys 7A80C8ED4FCCBE09

I suspect both keys are valid, but in either case navigating this step was fairly straightforward and the prompts made executing the command a matter of cutting and pasting.

Having verified the source, I was able to then update apt-get and install graph-tool for Python 3:

apt-get update && \
apt-get install python3-graph-tool

Once install, I was able to confirm that the files were present, under /usr/lib/python3/dist-packages/. Because the rest of the Python 3 files were located under /usr/local/lib/python3.6/site-packages (see below for what current paths were), I need to add to the available system paths the new path (under /python3/).

$ python -m site
sys.path = [

At this point, I could drop into a Python 3 REPL and import most components of graph-tool. But, in order to render graphs, I needed a few more libraries, specifically:

  • PyCairo, a Python-bindings library for the Cairo graphics tool (so also Cairo)
  • PyGraphviz a Python-bindings library for the Graphviz network dot graph rendering tool (so also Graphviz)

I believe Cairo was already installed but, regardless, it’s fairly straightforward to install:

apt-get install libcairo2-dev

And the Python bindings were installed easily with pip:

pip install pycairo

The peartree Makefile already contains a script for installing Graphviz and its Python bindings. Its installation is also straightforward:

    apt-get update && \
    apt-get install xdg-utils && \
    apt-get install graphviz && \
    pip install pygraphviz


At this point, the peartree development environment has been “upgraded” sufficiently to support operation of graph-tool (and rendering of graph components). Hopefully this documentation is not just helpful in documenting my efforts with regards to peartree, but also others attempting to install graph-tool on their own in the future.