Backwards-compat breaking changes

[falexwolf]

A list for keeping track of things that we might change when breaking backwards compat at some point:

• merge sparse pca
• merge Change the default number of nearest neighbors search in Ingest #1111
• merge Add 'equal_frequency' option to highly_variable_genes #572
• set the cachdir default to user_cache_dir(…), ~/.scanpy/cache/ or ~/.cache/
• stationary states in DPT: https://github.com/theislab/scanpy/blob/b11b4abe5e16053c010e57b2dd3a27396a4b0cf2/scanpy/neighbors/__init__.py#L853-L857 thanks to @Marius1311 for pointing it out!
• rename log2fc or similarly: Documentation for rank_genes_groups  #446
• add inplace functionality where easily possible, that's not a simple renaming; a function that has inplace in it, should only return the annotation if inplace=False; the copy functions return the whole adata, which we don't want...
• rename n_comps to n_components everywhere
• replace default pca solver with 'arpack'
• change default solver in logreg solver in rank_genes_groups to lbfgs
• merge Fix t-tests when variance is zero #621
• Transition away from positional APIs: Transition away from positional APIs #464 (actually backwards compatible through decorator!)

done in #3653

• make t-test or wilxocon the default of tl.rank_genes_groups
• make pp.highly_variable_genes return a df instead of a recarray...

anndata:

• merge Maintain dtype during initialization anndata#130 and fix Scanpy tests

[ivirshup]

Could this issue include the previously discussed (#316 (comment)) change from copy to inplace?

[flying-sheep]

sure. could you please edit your comment to contain a link to said discussion?

[ivirshup]

I've edited my comment. I'd also be very in favor of the differential expression API getting reworked (#61). It's possible diffxpy should cover this, but I'm not really sure what the state of its development is.

[falexwolf]

Completely agreed, @ivirshup!

The inplace and copy change is not just about renaming. See my edit of the check box list above.