DOC custom scoring usage GridSearchCV and RandomizedSearchCV

[siddu1324]

Reference Issues/PRs

References #28671

What does this implement/fix? Explain your changes.

This PR adds documentation examples for using custom scoring functions with GridSearchCV and RandomizedSearchCV, specifically illustrating how to use make_scorer for metrics requiring additional parameters, like d2_pinball_score. This enhancement addresses user requests for clearer guidance on applying custom scorers in model selection.

[github-actions]

✔️ Linting Passed

All linting checks passed. Your pull request is in excellent shape! ☀️

_{Generated for commit: 039d1b1. Link to the linter CI: here}

[jeremiedbb]

Thanks for the PR @siddu1324. I'm not sure the notes is the right place for this added doc. I don't think it would have helped figuring out how to correctly use the scoring parameter.

I think improving the scoring parameter description would have a better impact. ping @ogrisel who originally answered in the linked issue, wdyt ?

[ogrisel]

In addition to @jeremiedbb's remark above about the location of the snippet in docstring, there are several problem:

• the linear regression model does not accept an alpha parameter. Calling fit on a dataset generated by make_regression would raise:

ValueError: Invalid parameter 'alpha' for estimator LinearRegression(). Valid parameters are: ['copy_X', 'fit_intercept', 'n_jobs', 'positive'].

• I would also rather not tune a hyperparameter that has the same name as the metric parameter to avoid introducing any confusion;
• furthermore, it's weird to tune a linear regression model that estimates the expected value of the target variable conditionally on the features on a metric that assess it's ability to estimate a 0.95 quantile. I would instead a quantile estimator for this loss or alternatively use another parametrized metric such as fbeta_score on a simple classifier such as LogisticRegression.

[ogrisel]

Also this is the docstring of the GridSearchCV class but this code snippet shows how to use the RandomizedSearchCV instead.

A similar can be added in the inline examples section of each of those classes but should be adapted accordingly.