docs: Add a technical example showing scikit-learn API compatible estimators that work with skore

[sylvaincom]

Note: following #1814 (comment) in June 2025, this issue has broadened its scope from just deep learning to any estimator compatible with the scikit-learn API

Which part of the documentation needs improvement?

Examples gallery

Describe the problem found in the documentation

No deep learning methods, no foundations models, etc are mentioned in our examples, mainly traditional scikit-learn models, while skore can deal with basically any estimator that is compatible with the scikit-learn API.

Suggested improvement

Add an example in the technical section

Its goal is to explain that you can pass to skore any model that has a predict method; and use skorch, keras (see keras-team/keras#20599), tabpfn, etc, a manual wrapped model and explain a bit the internal details to show when you will get some guidance with skore (as the choice of metrics)

Additional context

Could also do a wrapper for a LLM, as done in https://medium.com/capgemini-invent-lab/quantifying-llms-uncertainty-with-conformal-predictions-567870e63e00

[MarieSacksick]

@warith-harchaoui, @emmatysinger, would you be willing to take this one? Should we plan a discussion about it?

[sylvaincom]

I should have specified that we had a meeting about this with Warith and Emma on March 6th, this GH issue is kind of like a placeholder

Minutes on Notion: https://www.notion.so/probabl/Brainstorm-Deep-Learning-example-for-skore-1aeef76d36b980488845dc41b4e8b089?pvs=4

[MarieSacksick]

I heard about it, but the output is not clear to me; should this be done, and if yes, by whom?

[sylvaincom]

This should be done, I'll try to do it, but it'll take me a few half-days I believe, so depends on the priority we want for this

[nkapila6]

Hi @sylvaincom, ref our discussion at #1557

I don't have access to the minutes of meeting but looking at your code here, it seems like you're reporting top-k scores for k=1,5,10 using skore on a pre-trained ResNet.

Would you like me to implement the same using skorch?

[sylvaincom]

Hi @nkapila6, thanks for following this up. No need to have access to the minutes, we were just discussing potential deep learning use cases. Sure, you can try something with skorch. Beware of the memory usage and processing time: the dataset must be not too large but still interesting (not a synthetic dataset) and the example is actually executed on our documentation so must take a few minutes to run. Let me know if you need any more information. You can also directly try something on the same datasets used in the examples of skorch, but applying skore on it.