Towards a community for open-source software in sports

#THIS ARTICLE IN DUTCH is written by Leo Aquina | translated by Liesbeth Wallien

In 2023 data have become indispensable in sports. But software engineer Koen Vossen is convinced there is still much to be gained in terms of data analysis used to improve performance in sports. The more people cooperate the better the results. This is why he is keen to encourage the use of so-called open source software. Analogous to PyData, well-known in IT circles, he founded PySport, which should become a community sharing and further developing open-source software in sports. As the founder of TeamTV Vossen has worked for over ten years on software for video analysis for clients in different sports.

‘I have been programming since I was twelve years old and once I graduated from high school I started working as a web developer right away’, Vossen says. ‘All my life I have played korfball and I am a trainer/coach as well.’ When in 2013 he started TeamTV with fellow korfball coach Ron Steenbergen, he had the chance to combine both passions. ‘We make sure that sport clubs and competitions can use video, live streams and data for analyses as efficiently as possible. Meanwhile we have clients in football, handball, korfball and basketball.’

Python
As a web developer Vossen is a member of PyData. ‘It gets its name from Python, a language used a lot in data science’, he explains. ‘PyData is a project of NumFOCUS, an international not for profit organization that supports open-source software. Under the name PyData they organize conferences worldwide.’ Vossen was interested specifically in sport software and discussed this with NumFOCUS. ‘They were willing to support it, but sports is a comparatively small area of focus, which is not of interest to the major part of PyData’s other activities. That’s why we created PySport, which is a separate organization. Officially it is not linked to NumFOCUS, but we would like a similar organization. In April we held a first meetup and we are now exploring whether we can organise a conference, possibly to generate income that, like NumFocus, we could use for open-source projects.’

One example of such an open-source project is Kloppy, which Vossen himself launched. ‘I noticed that people were sharing code examples, but eighty percent of that code consisted of reading in files that came from sources like Opta or other parties. These are files that record what happened, for instance at a football match, but they all have different formats. I felt it was a waste that people wanted to share good ideas but had to spend 80% of their time reading in those files. With Kloppy those files can be standardised in an Excel format, regardless of the source, which means they are easier to use by people in the sports data community. With Pysport I want to bring those packages to the attention of users as well as developers, so people don’t have to reinvent the wheel.’

Revenue model
But why would people using data to improve sport performance want to share their knowledge? Every sportsman and woman, after all, wants to get a competitive edge over their competitors? Vossen: ‘That is indeed the reasoning of the clubs. Of course everyone decides what they do and what they don’t share. But with a technique like Kloppy that makes it easier to compare data side by side, all clubs benefit without sharing competitive information.’ What is left of the revenue model of parties for whom sports data are big business, like Opta, if data analysist use open-source software? Vossen: ‘Parties developing software in house should make sure that whatever they have to offer outdoes the available open-source software. They can do so by building on existing open-source software, or by adding consultancy services. It often takes specific knowledge to use it well.’

The first PySport-meetup in Eindhoven attracted all sorts of data scientists from different sports. Under the heading 'Adding context to football: How to implement a playing phase model in Python' Rogier Noordman of PSV explained how to use Python to make a model of game phases based on game principles. Vossen himself talked about the possibilities of Kloppy, and Joris Bekkers of UnravelSports held a presentation about the techniques behind data visualization.’

For Vossen the Eindhoven meetup was just a start. ‘We began with building an online community. Currently we have some 5,500 followers on Twitter and 1,400 on LinkedIn. We have two valuable online sources for sports analysts and anyone who is interested. On the PySport website users can find an overview of the open-source packages available for sports analysis and in the PySport playground users can experiment with several of those open-source packages in a quick and easy way.’ Vossen has meanwhile been invited by the German football federation DfB to organise a meetup at their offices, and from Belgium and Poland, too, he got enthusiastic responses. ‘We hope to organise the first conference after a couple of meetups, but we need an underlying organisation in the form of a foundation. That’s what we are currently working hard on.’

For more information: opensource.pysport.org