docs(getting started): Change dataset to synthetic

[mohsinm-dev]

Summary

• Replace load_breast_cancer with make_classification using challenging parameters
• Creates a more compelling demonstration of Skore's capabilities
• The breast cancer dataset produces near-perfect results that don't effectively showcase the model
  evaluation tools

Changes

• Use make_classification with parameters: n_samples=1000, n_features=20, n_informative=10,
  n_redundant=10, n_clusters_per_class=1, random_state=42
• Convert to DataFrame format to maintain compatibility with existing code
• Update comments to reflect synthetic dataset usage

Why this improves the example

The synthetic dataset provides more realistic performance metrics (ROC-AUC ~0.90 vs ~0.99) and
clearer differences between models, making it better for educational purposes, and demonstrating
When different models/techniques matter.

Test plan

• Verified EstimatorReport works with the synthetic dataset
• Tested CrossValidationReport functionality
• Confirmed ComparisonReport works correctly
• Validated feature importance calculations
• Ensured all metrics and visualizations display properly

Fixes #1861

[thomass-dev]

Did you use an LLM to generate your PR? Please check the documentation.

The challenge is to understand the issue, and to be able to easily contribute in future.
Thanks.

[thomass-dev]

I don't think the PR is sufficient to resolve the issue. @sylvaincom can you take a look? Thanks!

This is the current rendering:

[github-actions]

Documentation preview @ d973f3e

[sylvaincom]

Hi, I agree with @thomass-dev: this feels LLM generated and you did not take into account the description of the parent issue, especially the shown plots

comparator.metrics.roc().plot() should display a ROC curve that does not appear too easy

[mohsinm-dev]

@sylvaincom I’ve updated the example to use 10,000 samples with CrossValidationReport and matched the code structure to your reference. Can you please check this ?

[sylvaincom]

You do not seem to understand what you're doing: you're calling rf_report (RF : random forest) a model report corresponding to a logistic regression

[glemaitre]

A couple of remarks.

[auguste-probabl]

Here's what you get with RandomForestClassifier:

CVReport:

ComparisonReport:

IMO it looks good enough for our purposes.

[thomass-dev]

Thanks @auguste-probabl for the follow-up, LGTM now.