Adjust regularization and data pre-processing to ensure logistic regression converges

[david-cortes-intel]

Description

Many benchmark cases for logistic regression are executing something that does not reach convergence within the specified tolerances. This PR adjust the cases used for logistic regression so as to use a more appropriate regularization value for each dataset, and pre-process them in ways that would avoid ending up with features in too different scales.

Before:

After:

---

Checklist:

Completeness and readability

• Git commit message contains an appropriate signed-off-by string (see CONTRIBUTING.md for details).
• I have resolved any merge conflicts that might occur with the base branch.

Testing

• I have run it locally and tested the changes extensively.
• All CI jobs are green or I have provided justification why they aren't.

[david-cortes-intel]

CI error is from an xgb model conversion issue:

                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/share/miniconda/envs/bench-env/lib/python3.11/site-packages/daal4py/mb/gbt_convertors.py", line 546, in get_gbt_model_from_xgboost
    base_score = float(xgb_config["learner"]["learner_model_param"]["base_score"])
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ValueError: could not convert string to float: '[-2.5816395E0]'

It should be fixed with this PR in sklearnex: uxlfoundation/scikit-learn-intelex#2741

[Vika-F]

The changes look good to me.
The only question I have: Will it be possible to merge the resulting jsons after these changes with the jsons collected with the current version of the benchmarks? Or they would need to be recollected with the updated version?

[david-cortes-intel]

The changes look good to me. The only question I have: Will it be possible to merge the resulting jsons after these changes with the jsons collected with the current version of the benchmarks? Or they would need to be recollected with the updated version?

They would be mergeable, but there would be fewer overlapping entries so it wouldn't show comparisons for most cases.