[python] CensusSCVIDataModule + notebook

[ebezzi]

This PR contains:

1. An implementation of CensusSCVIDataModel that allows to train an scVI model using the Pytorch data loaders
2. A notebook that explains how to do the training end to end. Preview here

Note for reviewers: marking as draft as there are some items missing:

1. Still depends on [python] Support custom obs encoders #1191, which in turns require some testing/changes. Currently WIP
2. CensusSCVIDataModule needs to be refactored to comply with our linter

[pablo-gar]

@ebezzi Should we add torchdata as a dependency of the experimental package? it's required by CensusSCVIDataModule

[pablo-gar]

I can't reproduce the notebook. @ebezzi do you have any hints as to why?

https://github.com/chanzuckerberg/cellxgene-census/pull/1196/files#r1660286829

[pablo-gar]

Can we move all imports here?

[pablo-gar]

@ebezzi can you please add two cells at the end:

sc.pl.umap(adata, color="tissue_general", title="SCVI")

and

sc.pl.umap(adata, color="cell_type", title="SCVI")

[ebezzi]

@ebezzi Should we add torchdata as a dependency of the experimental package? it's required by CensusSCVIDataModule

It's already in the requirements (see pyproject.toml)

[codecov]

Codecov Report

Attention: Patch coverage is 92.59259% with 6 lines in your changes missing coverage. Please review.

Project coverage is 91.16%. Comparing base (8d1e103) to head (6edd123).
Report is 30 commits behind head on main.

Files	Patch %	Lines
...us/src/cellxgene_census/experimental/ml/pytorch.py	90.62%	6 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1196      +/-   ##
==========================================
+ Coverage   91.11%   91.16%   +0.04%     
==========================================
  Files          77       77              
  Lines        5923     6022      +99     
==========================================
+ Hits         5397     5490      +93     
- Misses        526      532       +6     

Flag	Coverage Δ
unittests	91.16% <92.59%> (+0.04%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

This PR has not seen any activity in the past 4 weeks; if no one comments or reviews it in the next 3 days, this PR will be closed.

[pablo-gar]

Comment to keep PR open