Dataset download resolutions

[ethanglaser]

Description

Addresses timeouts and problematic dataset downloads in CI jobs by:

1. Replacing openml usage from sklearn.datasets.fetch_openml to using openml package directly
2. Removing epsilon dataset

---

Checklist:

Completeness and readability

• I have commented my code, particularly in hard-to-understand areas.
• I have updated the documentation to reflect the changes or created a separate PR with updates and provided its number in the description, if necessary.
• 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.
• I have extended testing suite if new functionality was introduced in this PR.

[ethanglaser]

Latest run with CPU: http://intel-ci.intel.com/f1085339-fdd4-f15e-a36a-a4bf010d0e2d

[ethanglaser]

Rerun: http://intel-ci.intel.com/f1279be9-cb88-f16a-b6b1-a4bf010d0e2d

[ethanglaser]

/azp run CI

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[david-cortes-intel]

@ethanglaser The CI error:

ModuleNotFoundError: No module named 'openml'

Probably you need to add it to this file:
https://github.com/IntelPython/scikit-learn_bench/blob/main/envs/conda-env-sklearn.yml

[david-cortes-intel]

The CI error:

INFO - sklbench - Report summary
Empty DataFrame
Columns: []
Index: [ElasticNet|fit, ElasticNet|predict, KMeans|fit, KMeans|predict, KMeans|transform, KNeighborsClassifier|fit, KNeighborsClassifier|predict, KNeighborsClassifier|predict_proba, PCA|fit, PCA|transform]

Not sure what's causing it though.

[david-cortes-intel]

Actually this is the error:

ImportError: cannot import name '_check_multi_class' from 'sklearn.linear_model._logistic' (/usr/share/miniconda/envs/bench-env/lib/python3.11/site-packages/sklearn/linear_model/_logistic.py)

But again not sure what's the issue - I see it's using sklearn1.8 with sklearnex2025.11, which should be compatible.

[david-cortes-intel]

Digging a bit into it, I think the issue is with this particular line in sklbench:

scikit-learn_bench/sklbench/utils/common.py

Line 157 in 399d9eb

module = importlib.__import__(module_name, globals(), locals(), [], 0)

It appears to be importing submodules that estimators use into global variables.

Since daal4py defines a module sklearn, it imports that as "sklearn", and then later on when it comes the time to import things from scikit-learn which shares the same module name, it leads to clashes as it tries to look for functions from one in the other.

Apparently the only use for that function is to determine if a particular library contains a given estimator or function by name:

scikit-learn_bench/sklbench/benchmarks/sklearn_estimator.py

Line 54 in 399d9eb

if estimator_name not in classes_map:

scikit-learn_bench/sklbench/benchmarks/custom_function.py

Line 33 in 399d9eb

if function_name not in functions_map:

Perhaps a quick (but very inefficient) fix could be to do those checks in a subprocess or in a forked process. Actually it's also used to return the estimator class, so it cannot be put into a separate process.

[david-cortes-intel]

LGTM, issue with the failing jobs is not related to these fixes and could be left for a later PR.