Describe the issue linked to the documentation

While working on a docstring-related pull request (#24259) I noticed that, sometimes, the See Also description for the same function/class does not match. For instance, the accuracy_score description was different depending on the class I looked at:

I decided to investigate it further and see if it was occuring elsewhere (click here to check the gist where I regexed the raw files). Today, at commit 2f8b8e7f1, we have 56 functions/classes that have some sort of different descriptions at some See Also. From looking at them at first glance I have sorted the differences I get into 4 categorys:

1. Those where the description is related to the current class/function;

   • For instance, inside the RegressorChain class, ClassifierChain is described as the "equivalent" version. That is a different description from the one we see inside MultiOutputClassifier, but inside the context of the RegressorChain class, it makes sense.
     

     scikit-learn/sklearn/multioutput.py

     Lines 965 to 967 in 2f8b8e7

     	See Also
     	--------
     	ClassifierChain : Equivalent for classification.

2. Those related to extra "\n" on the text;
   

   scikit-learn/sklearn/kernel_approximation.py

   Line 106 in 2f8b8e7

   SkewedChi2Sampler : Approximate feature map for "skewed chi-squared" kernel.

   
   

   scikit-learn/sklearn/kernel_approximation.py

   Lines 292 to 293 in 2f8b8e7

   	SkewedChi2Sampler : Approximate feature map for
   	"skewed chi-squared" kernel.

3. Those that are different per se;

   • For instance, the accuracy_score descriptions we mentioned earlier have no obvious reason to be different.

4. Those that are related to the docstring pattern.

   • For instance, we are missing a space after the name of the class/function being referenced. So my script puts it inside the previous description (extra question: shouldn't it be failing on the numpy doc validation? I was excepting it to be an open docstring from Ensure that functions's docstrings pass numpydoc validation #21350, but this function was already done by DOC Ensures that dcg_score & roc_curve passes numpydoc validation #24351).
     

     scikit-learn/sklearn/metrics/_ranking.py

     Line 961 in 2f8b8e7

     det_curve: Compute error rates for different probability thresholds.

Suggest a potential alternative/fix

I was planning on taking a closer look at the output table I get with the differences in the descriptions and coming up with the list of functions/classes that we should change to make it uniform across different occurrences of See Also.
Then, as this is a fairly beginner friendly issue, I would edit the issue description and give it a step-by-step (copying most part of what @thomasjpfan did in #21350) and let folks looking for a "good first issue" help. I would be happy to make sure the task bullets are up to date.

What do the maintainers think? Does it make sense? :)

PS: Of course, we will be forever vulnerable to this kind of issue if we don't create a test (maybe adapted from my gist). The problem here is that sometimes it just makes sense to have different descriptions... We can create an "ignore list" maybe, but we can discuss this later.