Dimred datasets preprocessing

[lazappi]

Submission type

This submission modifies the dimensionality reduction (2D) task to include pre-processing steps in the data loaders (See #279). These should store a higher dimensional embedding in adata.obsm['X_input'] which is then used as the input for (most) methods.

• Add a preprocessing.py function that has pre-processing functions (one currently)
• Added that pre-processing function to the existing data loaders
• Modified the existing methods to use the pre-computed higher-dimensional embedding as input
  • PHATE is an exception, wasn't sure if that was appropriate so skipped for now but should probably discuss before merging
• Updated the API and README to describe the change

Testing

• This submission was written on a forked copy of SingleCellOpenProblems
• GitHub Actions "Run Benchmark" tests are passing on this base branch of this pull request (include link to passed test: ) PENDING
• If this pull request is not ready for review (including passing the "Run Benchmark" tests), I will open this PR as a draft (click on the down arrow next to the "Create Pull Request" button)

Submission guidelines

• This submission follows the guidelines in our Contributing document
• I have checked to ensure there aren't other open Pull Requests for the same update/change

[lazappi]

Thinking about this a bit more I'm actually wondering if it is better to do this on the methods level rather than the datasets level? We could still have a preprocessing.py file with the functions but call them as part of a method rather than a data loader.

CONS: Pre-processing would be run multiple times (maybe bad if there are many methods), more work to add methods
PROS: More flexible in which methods can handle which pre-processing, less work to add datasets

The results at the end should be the same so I guess it depends which the core members would prefer and is most consistent with what is being done for other tasks.

[cramirezs]

Thinking about this a bit more I'm actually wondering if it is better to do this on the methods level rather than the datasets level? We could still have a preprocessing.py file with the functions but call them as part of a method rather than a data loader.

This is what I was referring to with "... we just put the preprocessing in each method version as sqrt_umap, log1p_umap, etc." in #279. I guess it depends on how costly this step will be. I am not sure if talking about parallelization makes sense (I haven't used AWS).

[LuckyMD]

So far we have converged on preprocessing as part of methods if possible in openproblems.

[lazappi]

Based on @LuckyMD's comments in a couple of places it seems like moving the pre-processing to the method layer would be more consistent with other tasks.

I will close this now, move some things around and start a new PR.