EuroSciPy 2025 Array API Standard Debrief

[lucascolley]

EuroSciPy 2025 felt successful for the standard, and to a surprising degree!

My talk which gave an overview of the standard’s ecosystem was very well attended, with lots of questions at the end, and people coming to ask more in the days which followed. @paddyroddy presented a poster about adopting the standard in https://github.com/glass-dev/glass and had quite a few people coming to ask questions about the experience. @seberg mentioned the standard in his talk on delegation/dispatching more broadly, and was also answering questions about the standard to various people.

@FrancescAlted and @lshaw8317 from https://github.com/Blosc/python-blosc2 mentioned that they are planning to prioritise support for the standard, and were working on it during the sprints (Francesc also opened #968). Overall, between people working on array-api-extra or SciPy¹ (with me), scikit-learn (with @ogrisel), and their own projects, I think we had about 10 people sprinting on the standard in some form — which is a lot, considering there were only 130 people in total at the conference, and maybe about half as many still there for the sprints! @brisvag from Napari was investigating supporting the standard in one of his projects (maybe Napari itself?), and @OleBialas was working on support inside https://github.com/powerfulbean/mTRFpy.

Multiple people mentioned the standard feeling like a big theme of the conference, including @jacobtomlinson in his LinkedIn post. I honestly wasn’t expecting to see this much interest in and enthusiasm for the standard, but I think it is evidence that what we are building is serving real interests within the community, in a way which people understand. It was really nice to see things coming together and to feel the momentum building — it has left me excited to see how things will continue to grow in the future!

Just a shame that not more members of the consortium could make it — next year @ev-br !

Now onto PyData Paris to do the same again 🥖🍷🧀

Footnotes

1. https://github.com/rgommers/pixi-dev-scipystack was also a great success by the way, with @VeckoTheGecko able to run the tests across all array backends as a first time contributor without needing to wrangle with any complicated instructions! ↩

[FrancescAlted]

FWIW, besides Sebastian Berg's reference to array API in his talk, an eye-opener to me has been the poster by you, @lucascolley, about masked arrays for all packages that follow the array API. I genuinely think this could open the door to a new era where data containers will provide what they do best (for Blosc2, compression, computation and remote access), and then leave many advanced functionality (like masked arrays) that can work on top of data containers that are array API compliant, without having to reinvent the wheel.

[lucascolley]

Ah yes, I should have mentioned that also! Credit goes to @mdhaber for the vast majority of the work behind that poster. We should try to get Blosc working with MArray soon!

[betatim]

Nice work!

[rgommers]

Thanks for sharing @lucascolley, looks like excellent momentum and a fun conference!