Description
Hi all 👋
I hope you're all well! I'd like to discuss what should happen to this repository and project going forwards
I think the Consortium had some good ideas, and I appreciate the efforts to improve tool composability. I have a lot of respect for everybody involved. We just need to recognise that activity has completely stalled, and the project seems to have been abandoned
This repository has been the home to two projects:
- Dataframe interchange protocol. This hasn't been updated in 1.5+ years, and design issues haven't received responses nor fixes. I find this a bit disappointing, as a lot of code was submitted to pandas and maintenance then fell entirely on pandas' shoulders. Anyway, that ship has sailed, let's focus on deciding what happens next 🤝
- Dataframe API standard. This never took off - it's a pity, because as it turned out there was quite a bit of demand for this, we just couldn't agree on fundamental design issues.
Regarding adoption of the former:
- cuDF have removed support for it completely
- I've moved pandas and Polars
from_dataframeover to try using the PyCapsule Interface first. I hope nobody holds this against me, I don't think there was any other option - Consumers have generally moved away from it (either towards the PyCapsule Interface or towards Narwhals, or both in the case of Vegafusion). Scikit-learn use it for
.column_namesand for checking if a dtype is categorical, but that's about it, and those checks are easily replaceable
At this point - what next? To reduce fragmentation and avoid disappointing people, I'd like to suggest that this repository be archived and/or be marked as deprecated, possibly along with some statement about lessons learned?
I hope this message comes across well, and that you're all doing well ☺
Thank you for your consideration 🙏