Fix differing versions of dependency scikit-learn

[ckunki]

Solution strategy

Add cells to AI-Lab notebook nbtest_sklearn.py

• Detect scikit-learn versions inside the database SLC and in the AI-Lab
• If different then call Jupyter magic pip to replace the AI-Lab version with the one used in the SLC:

%pip install "scikit_learn=={slc_version}"

Notes

• This requires pip to be available in the PATH, see Add jupyterenv to PATH  #273
• This should also enable fixing the notebook tests as these can run all cells in a notebook including the cell fixing the scikit-learn version

Sample Snippets

Checking the version of sklearn in a shell

python -c "import sklearn; print(sklearn.__version__)"

Installing sklearn:

python3 -m pip install "scikit_learn==1.4.2"

Sample UDF for detecting scikit-learn version inside the database SLC

--/
CREATE OR REPLACE PYTHON3 SCALAR SCRIPT detect_skikit_learn_version() RETURNS VARCHAR(100) AS
import sklearn
def run(ctx):
    return sklearn.__version__ 
/

select detect_skikit_learn_version();

Background

During implementation of #301, PR #311 showed an error when running nbtest_sklearn.py

FAILURES ... test_regression[StorageBackend.saas]
AttributeError: 'OneHotEncoder' object has no attribute '_infrequent_enabled'

Stage	scikit-learn version	cloudpickle
SaaS prod	1.0.2	not installed
SaaS staging	1.4.2	installed
Docker DB	(unknown)	(unknown)

Other options

• use ONNX, PMML as exchange format, see https://scikit-learn.org/stable/model_persistence.html
• use pure Json or similar: high effort requires manually instantiating all scikit-learn, depends highly on version of scikit-learn

[ckunki]

The following zip contains an SQL script for running pip and the results of pip list on different stages.
sql.zip