feat: Enhance skore.cross_validate

[sylvaincom]

skore.cross_validate`

First PR of this new feature: #443, cc @MarieS-WiMLDS and @augustebaum. Issue: #383.
Some issues arose in #443 (comment), that have been separated to this new GitHub issue. The issue is written again here:

---

• feat: Enhance skore.cross_validate with more readable time #564
• feat: Enhance skore.cross_validate with average and standard deviation #562 Add the average and std on the plots, as done in the scikit-learn MOOC:

EDIT : after a discussion with @augustebaum, maybe we can keep both, but in all cases, we should say somewhere what is actually behind test_score.

For the fit time:

• We must specify the units (seconds for example) of the fit time
• We must distinguish scores (the higher, the better) and errors (the higher, the worse), for example this is an error but the title is still "Cross-validation scores" and the y-label is "Score": this is misleading

Note: neg_mean_squared_error is actually meant to be negative and behave like a score (the higher, the better). But still, naming "results" is better I believe.

Maybe we can say something like "Cross-validation results" and the y-label is "Results", so that it can be appropriate for scores, errors, and times

EDIT by @augustebaum : Done

• We should switch to plotly before the first release, because DS are much more familiar with plotly than with altair. See dropdowns in plotly.

EDIT: Tis is the new display that does not have a dropdown menu but does the same & it can enable us to compare the training and testing times together which is very useful ; if we had the train score then we can also compare the train and test scores together

Enregistrement.de.l.ecran.2024-10-17.a.17.46.57.mov

[MarieS-WiMLDS]

Another short enhancement:

• feat: Enhance skore.cross_validate with cross_validate history #563 do not erase cross_validate on each run, but rather create a new plot with a new name, following this convention: cv_skore_{estimator}run{iter}

[augustebaum]

To be honest, this issue is getting a bit big. Can we create individual issues?

[MarieS-WiMLDS]

@augustebaum I created 3 smaller issues.
@sylvaincom, I didn't create issues about For linear regression, it is redundant to say test_r2 and test_score. because I don't know what kind of design solution you had in mind, nor for the train scores, for the same reason. I'll let you do it :)!

[sylvaincom]

Created to two issues of this larger issue that I would consider a small epic

[augustebaum]

Now that this has been divided up I'll go ahead and close this