E IndexError: list index out of range on Python 3.13

[Zethson]

Describe the issue:

Minimal Complete Verifiable Example:

import dask_ml.preprocessing as daskml_pp

leads to

/opt/hostedtoolcache/Python/3.13.1/x64/lib/python3.13/site-packages/ehrapy/preprocessing/_normalization.py:13: in <module>
    import dask_ml.preprocessing as daskml_pp
/opt/hostedtoolcache/Python/3.13.1/x64/lib/python3.13/site-packages/dask_ml/preprocessing/__init__.py:5: in <module>
    from .data import (
/opt/hostedtoolcache/Python/3.13.1/x64/lib/python3.13/site-packages/dask_ml/preprocessing/data.py:1048: in <module>
    class PolynomialFeatures(DaskMLBaseMixin, sklearn.preprocessing.PolynomialFeatures):
/opt/hostedtoolcache/Python/3.13.1/x64/lib/python3.13/site-packages/dask_ml/preprocessing/data.py:1060: in PolynomialFeatures
    __doc__ = "".join([splitted_orig_doc[0], __doc__, splitted_orig_doc[1]])
E   IndexError: list index out of range

Anything else we need to know?:

Works flawlessly in my CI with Python 3.12

Environment:

 + dask-glm==0.3.2
 + dask-ml==2024.4.4

• Dask version: 2025.1.0
• Python version: 3.13
• Operating System: Linux
• Install method (conda, pip, source): uv pip

[TomAugspurger]

Thanks for the report. Can you confirm whether or not you have the same version of scikit-learn in those two environments?

[TomAugspurger]

Ah, this seems to be from the first bullet at https://docs.python.org/3/whatsnew/3.13.html#other-language-changes

The compiler now strips common leading whitespace from every line in a docstring.

We'll need to make the docstring handling more robust. str.partition might be better here than str.split.

[TomAugspurger]

Should be fixed by #1008.