First try at deg pred

[PierreBoyeau]

Fixes #20

I tried something for the DEG prediction,

There are two components.

1. The main distance computation can return a tensor of LFCs of shape (n_samples, n_samples, n_genes). Element i, j, g computes the LFC for gene g between counterfactuals for samples i and j, aggregated over all cells.
2. This output can be used to rank genes given some binary sample metadata. To do so, we could compute the mean of the LFC for pairs of donors s1, s2, s.t. s1 has metadata 1, s_2 metadata 2

I also computed a dataframe mapping scvi sample ids to metadata covariates.

[PierreBoyeau]

there is a bug in the computation of step 2, i will correct it ASAP

[codecov]

Codecov Report

Merging #25 (0db45d3) into main (1eced8f) will increase coverage by 0.19%.
The diff coverage is 98.11%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #25      +/-   ##
==========================================
+ Coverage   98.67%   98.87%   +0.19%     
==========================================
  Files           6        6              
  Lines         302      355      +53     
==========================================
+ Hits          298      351      +53     
  Misses          4        4              

Flag	Coverage Δ
unittests	98.87% <98.11%> (+0.19%)	⬆️

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

Impacted Files	Coverage Δ
src/scvi_v2/_model.py	99.29% <98.11%> (-0.71%)	⬇️
src/scvi_v2/_module.py	97.65% <0.00%> (+0.78%)	⬆️

[justjhong]

will this be closed for #34 ? @PierreBoyeau