In February 2025, I was part of the NetworkX leadership team that convened in Berkeley for a developers' retreat. We had three goals: to unblock some of our thorny development tasks, reflect on and update our long term vision, and spend time reconnecting in person.
Like many scientific open source projects, our team is distributed and primarily collaborates online. We coordinate work and engage community members around the world via our NetworkX GitHub repositories. We complement asynchronous communications with regular Zoom calls, which are open to the community.
Co-founder of the project and faculty in mathematics at Colgate University, Dan Schult, shared how important it is to bring the team together and create space for reflection and quick iterations on ideas and potential solutions:
Working together in person for a few days allows us to go back and forth intently about a topic, followed by an hour or two on something else, and then more about the first topic later when inevitably an idea comes to mind. The chance to follow up a short time later without waiting for another meeting is really helpful.
The positive effects of having the team focused on NetworkX for the same extended period of time were striking. During the retreat, we reviewed and merged 20 pull requests! The following week, we merged an additional 15.
NetworkX is a Python package for the creation, manipulation, and study of the structure, dynamics, and functions of complex networks. The open source code is downloaded more than 68 million times per month. Use cases for this tool include studying the dynamics of beliefs in populations, modeling power grid system failures, and mapping the human body at single cell resolution.
One of the notable pull requests we merged during the retreat is an algorithm for calculating an approximate solution to the densest subgraph problem: finding the subset of nodes that are most highly connected. This addition will significantly benefit researchers studying "communities" within networks by providing a more efficient way to identify densely connected node sets. For example, for many classes of graphs, this new algorithm is over 20 times faster than previous implemented algorithms.
Photo: Dan Schult, Jarrod Millman, Mridul Seth, Rick Ratzel and Stéfan van der Walt hiking in Berkeley during the NetworkX developers retreat in February 2025.
Rick Ratzel is a NetworkX contributor and technical manager at NVIDIA. His time and expertise benefit the NetworkX open source community by ensuring that the NetworkX API can be used on NVIDIA systems, accelerated by the cuGraph NetworkX backend.
At NVIDIA we’re supporting data scientists to speed up and scale their analyses with "zero code changes" required.
By contributing to the NetworkX community we support users to unlock large graph analyses on GPU hardware using the same code they have previously used on CPUs.
You can read more in Rick's tutorials, NetworkX Introduces Zero Code Change Acceleration Using NVIDIA cuGraph and Using NetworkX, Jaccard similarity, and cuGraph to predict your next favorite movie, on the NVIDIA developer blog.
Another focus of the retreat was to bridge the communications gap between users and developers of NetworkX. Unlike traditional software, open source software is obtained, with its source code, for free. And because there is no traditional provider-client relationship, it can be difficult for project leads to know whether their priorities align with user needs.
Photo: Jarrod Millman, Mridul Seth, and Dan Schult enjoying the beautiful Bay Area Hills while brainstorming the next stages of the NetworkX project vision.
Mridul Seth is a scientific software developer at the European Spallation Source. A long time contributor to NetworkX, Mridul led discussions around ways of gathering user feedback that the project can incorporate into their medium and long term plans.
We know we want user feedback, but we also have to grow our maintainer base to be able to respond to it. There's nothing worse than asking someone for their opinion and then not doing anything to address the challenges they've shared!
If you are using NetworkX or could benefit from the significant speedups provided by our collaboration with cuGraph, please share the feedback through our GitHub repositories or reach out to me directly. Additionally, we are always looking for collaborators on funding proposals. If you need support for graph analyses to achieve your scientific goals, please get in touch!
In summary, we held a successful NetworkX developers' retreat at BIDS in February 2025, marked by significant progress in development and a renewed focus on community engagement. As we move forward, we will build on this momentum by integrating user feedback into our development roadmap, updating tools for community detection and graph isomorphism, and expanding our collaboration with contributors like NVIDIA to enhance the capabilities of NetworkX.