Features for InterScale

[FrancescaDr]

Added possibility to add:

• var variables in addition to obs (to_category_iterator)
• one_hot_encoding for features to perform mapping back to feature names
• save preprocessed adata object for train, test and val

[selmanozleyen]

Hi thanks for the PR!

After removing the print statements and besides my comments I will still give an appoval to speed up the development on this project.

[selmanozleyen]

I like this idea but I think it might be better if we do it like this

preserve_categories: Optional[list[str]] = []

So empty list by default which would mean no preservation. And if it is None that would mean all categories. So then it would make sense to change the variable name to preserved_categories. Also to make the typing more modern finally it would look like this

preserved_categories: Sequence[str] | None = tuple()

here default is tuple instead because its not good to have default arguments as mutuable objects. Could you also update docstring with that?

[FrancescaDr]

Thanks for the review.

If we change preserved_categories to a tuple then how are we to distinguish between the "obs" and "var" variables?

[selmanozleyen]

But we specify if its a var or obs in axis right?

[FrancescaDr]

Yes true. I am now questioning whether this even works for .var. In the Ann2DataByCategory object the ToCategoryIterator is by default initialised with obs. Also in thetest_to_category_iterator() there is nothing tested for "var". Am I overlooking where the ToCategoryIteratorwould be initialised withaxis = var`?