DOC show from_estimator and from_predictions for Displays

[glemaitre]

closes #25980

I think a good enough solution for #25980 is to have a specific template for the displays.
It would give something like:

However, some displays do not implement from_predictions. Thus, an error will be raised by sphinx when we execute the include directive. I expect the CI to show an error then.

Does the Sphinx/HTML ninjas (maybe @lesteve @thomasjpfan) can think of some workaround to either include only if the file exists or discard those errors after sphinx compiling?

[glemaitre]

So we get somewhere in the log:

/home/circleci/project/doc/modules/generated/sklearn.inspection.DecisionBoundaryDisplay.rst:10: CRITICAL: Problems with "include" directive path:
InputError: [Errno 2] No such file or directory: 'modules/generated/sklearn.inspection.DecisionBoundaryDisplay.from_predictions.examples'.
/home/circleci/project/doc/modules/generated/sklearn.inspection.PartialDependenceDisplay.rst:10: CRITICAL: Problems with "include" directive path:
InputError: [Errno 2] No such file or directory: 'modules/generated/sklearn.inspection.PartialDependenceDisplay.from_predictions.examples'.
/home/circleci/project/doc/modules/generated/sklearn.model_selection.LearningCurveDisplay.rst:10: CRITICAL: Problems with "include" directive path:
InputError: [Errno 2] No such file or directory: 'modules/generated/sklearn.model_selection.LearningCurveDisplay.from_predictions.examples'.

However, it is not collected by bash and the CI does not fail.
I am wondering if it is not good enough then.

[lesteve]

Probably autosummary errors somehow don reach to the main sphinx process ...

This adds a bit of noise to the log which is slightly annoying in some cases e.g. when you are trying to figure out which warning could cause the build to fail. I would say this is good enough though.

A cleaner alternative would be to add a different template for classes that don't have from_predictions

[glemaitre]

A cleaner alternative would be to add a different template for classes that don't have from_predictions

This is a good idea. I should only check if we can easily switch templates in the same doc section

[jeremiedbb]

So we get somewhere in the log: [...]
However, it is not collected by bash and the CI does not fail.
I am wondering if it is not good enough then.

Is it because there are no examples using these ?

[glemaitre]

Is it because there are no examples using these ?

Not this is because this specific display does not implement from_predictions because it does not make sense.

[glemaitre]

So creating a new template is enough because the Displays are grouped properly.
Thanks @lesteve for the suggestion.

[jeremiedbb]

Now the question is can you inherit the template to avoid code duplication ? 😄
(I'm kidding, let's merge)

[lesteve]

FYI I don't think you need the full doc build to check that your PR works 😉 for example see this HTML page in your first commit where you can see your template is taken into account.

[glemaitre]

Now the question is can you inherit the template to avoid code duplication ?

I will implement this feature in the RST language directly :)