My question is both general and specific.

I was examining sklearn/naive_bayes.py while working on #15077 and #16281. The code was not particularly hard to read and follow, but I came up with some improvements without looking for them. For example, in _BaseDescreteNB I would:

• declare a couple of abstract methods
• rearrange code in fit and partial_fit and factor out a common part (just as it is done in GaussianNB)
• replace temporary variables with direct references to attributes, where it does not make any difference
• remove some shape checks, which were introduced 9 years ago before utils._validate_data was a thing.

In fact, I sort of "made" these changes while parsing the code, as they were part of the digesting. Since I've already put some effort into this, I was thinking, why not do a good thing by spending a few more hours (hopefully) and committing the changes and run some tests? I feel confident that overall my changes benefit readability and maintenance. However I have some concerns:

• I am not a professional software developer (yet, in any case), and refactoring may be not a task most suitable for novices.
• I figured out how things work, but I am not the one who wrote this code and cannot be 100% certain I haven't missed an important detail
• Therefore, there is possibility that the changes will crush someone's code while passing all sklearn tests (should I care?)
• Some of the changes might be a matter of taste
• None of the changes are, strictly speaking, necessary.

So I am asking for opinions and advice. How do people approach refactoring/restructuring/cleaning? Is there any sense in what I am saying? What approach and workflow do scikit-learn maintainers and developers follow?