FEAT: enhance scikit-learn's grid search

[sylvaincom]

Is your feature request related to a problem? Please describe.

As a user, I want to automatically get insights when performing the key machine learning task of grid search.

Describe the solution you'd like

After train-test split and cross-validation, the next ML diagnose we could tackle is grid search. Same as the CrossValidationReporter, we could have a GridSearchCVReporter that automatically shows a Parallel coordinates plot:

Enregistrement.de.l.ecran.2024-12-19.a.16.40.35.mov

Inspiration

• Sample pipeline for text feature extraction and evaluation -> see the Parallel coordinates plot towards the end
• hiplot -> can just be replaced by plotly's parallel coordinates plot as used in scikit-learn's example above
  • calmcode
  • Vincent D. Warmerdam - Run a benchmark they said. It will be fun they said. | PyData Amsterdam 2024 at 23:58 -> example of usage

API

Should be like CrossValidationReporter in philosophy, but here scikit-learn's grid search returns many things:

from skore import GridSearchCVReporter

param_grid = {
    "my_param_1": range(1, 10),
    "my_param_2": ["value_1", "value_2"],
}
grid_search_reporter = GridSearchCVReporter(
    my_estimator(), param_grid=param_grid, scoring=my_score
)
grid_search_reporter.fit(X)

my_project.put("gridsearch_reporter", gridsearch_reporter)
gridsearch_reporter.plots.parallel_coordinates

# like scikit-learn's grid search
print(gridsearch_reporter.cv_results_)
print(gridsearch_reporter.best_estimator_)
print(gridsearch_reporter.best_score_)
print(gridsearch_reporter.best_params_)

Other ideas for the grid search for next iteration

• For the GridSearchCVReporter, if only one hyperparameter, do the validation curve of scikit-learn
• Help to choose the range of values of a hyperparameter
• Comparing randomized search and grid search for hyperparameter estimation

Ready Checklist

• Design specification approved (by @MarieS-WiMLDS)
• Technical specification approved
• Critical considerations addressed
• (Optional) Planning requirements met