did you now decide to copy `skpro` instead of contributing to `skpro`?

[fkiraly]

Looks like it?

See older issue here: #1

Are you at least planning to cite skpro, or is ignoring prior art part of the general idea of copy-and-claim?

Are you planning to even acknowledge my contributions to the design of the scikit-learn compatible interface?

I guess not, right? Because academics will be academics.

[simon-hirsch]

Hi @fkiraly

Please reconsider your tone, I perceive it as quite aggressive and we encourage friendly discussions in ondil, even if we disagree.

did you now decide to copy skpro instead of contributing to skpro?

Short answer: No. Why do you make this assumption?

Long answer: No. I address your points one by one:

Looks like it?
See older issue here: #1
Are you at least planning to cite skpro, or is ignoring prior art part of the general idea of copy-and-claim?

No, we focus on distributional regression in the sense of modeling each distribution parameter as equation of explanatory features. We do need to implement distributions as we need the score functions and derivatives. All of this is missing in skpro. On the other side, we have no plans of wrapping other packages such as MAPIE or XGBoostLSS, providing pipeline functions or otherwise reproducing available functionality. If functionality is missing or needs a different design due to our online-first approach, we will implement it.

Online learning has been introduced in skpro in October 2024 (v2.7.0), after our talk at the sktime Meetup and after the publication of this package and paper. So much about acknowledging prior art (or my input to / comments on the skpro implementation sktime/skpro#462). There are many other online learning packages around (river, some MAPIE estimators etc), each with their special focus.

Are you planning to even acknowledge my contributions to the design of the scikit-learn compatible interface?

We value your input on the this early design change. We have acknowledged your contribution in the release notes (https://github.com/simon-hirsch/ondil/releases/tag/v0.1.9). We have a PR pending for acknowledging all contributors in a more prominent way in the README.md (see #155 and the branch readme).

I guess not, right? Because academics will be academics.

I perceive this as aggressive. Given the previous points, as well as the fact that all our communication so far has been characterized by mutual respect and interesting discussions, I don't get it and it's a bit sad.

We have plans to further contribute to community and provide adapters to sktime/skpro (see a draft PR here #152 where I played around with Agent Copilot). So #1 is still on our agenda. This kind of issue here is not helping the cause and is, at best, distracting. If you deem our progress to slow we much prefer to look at code contributions contrary to unfounded and angry issues.

I continue to see skpro/sktime and ondil as complementary libraries. I would be happy to strike a more positive note and further collaborate on the development of both packages.

[simon-hirsch]

Hi @fkiraly, are you considering to answer our comments? Otherwise I'll close this at the end of the week.