[WIP] Adding initial model based feature importance

[ctr26]

modified from EmbeddedArtistry

Description

Thank you for your contribution to pycytominer!
Please succinctly summarize your proposed change.
What motivated you to make this change?

Please also link to any relevant issues that your code is associated with.

What is the nature of your change?

• Enhancement (adds functionality).
• This change requires a documentation update.

Checklist

Please ensure that all boxes are checked before indicating that a pull request is ready for review.

Hi all,

I've unintentionally been working on my own similiar code base for years. If it's ok with you I'm going to upstream some of the code and modules that I think you're currently missing that would be helpful, particularly around extracting feature importances and model fitting. Hoping this is ok.

[gwaybio]

Thanks @ctr26 , we certainly welcome your contributions and we'd be delighted to include them!

We'll be happy to review this PR when it's ready, and we highly recommend adding functionality bit by bit. This will make review much more streamlined and effective.

I'll also provide two additional notes:

• please review our Contributing guide when you get a chance (if you haven't already of course!)
• we strive for high test coverage, so we are likely to be sticklers about testing

Cheers!

[d33bs]

Hi @ctr26 👋 — just checking in to see if you’re still planning to continue work on this PR. It's a great contribution and we’d love to help move it forward if you're still interested! If you need any help or context, feel free to ask.

If you’ve moved on or think someone else should pick it up, that’s totally fine too — just let us know what you’d prefer so we can figure out the next steps. Thanks again for your work on this! 🙌

[axiomcura]

A paper just came out:
https://www.nature.com/articles/s41592-025-02624-3

It presents a survey of 20 different feature selection methods for scRNA-seq data. I think we could potentially adapt some of these feature selection approaches for pycytominer.

However, it got me thinking... 💭 I think we need to clarify why we would incorporate these methods. Specifically, what is the justification for adding a new feature selection method? What objective metric or evidence do we have to demonstrate that a particular method is beneficial or necessary for pycytominer?

As far as I know, there isn’t a standardized way for evaluating and deciding when to introduce new methods in normalization, batch correction, or feature selection (i.e., what experiments or benchmarks are required). This might be a broader issue for contribution guidelines or issues templates? 🤔

Would it be useful to outline a set of questions or criteria that need to be addressed when proposing a new method?

Just wanted to leave my two cents here c: