De delta eps

[PierreBoyeau]

Fixes #1042

[codecov]

Codecov Report

Attention: Patch coverage is 93.61702% with 3 lines in your changes missing coverage. Please review.

Project coverage is 90.71%. Comparing base (0f860bf) to head (51dcc8b).

Files	Patch %	Lines
scvi/utils/_differential.py	93.61%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1043      +/-   ##
==========================================
+ Coverage   90.70%   90.71%   +0.01%     
==========================================
  Files          90       90              
  Lines        6615     6658      +43     
==========================================
+ Hits         6000     6040      +40     
- Misses        615      618       +3     

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

[adamgayoso]

maybe should make this percentile an argument with default value

[PierreBoyeau]

Do you think this should be a parameter in get_bayes_factors? I replaced this default value in estimate_pseudocounts_offset but not in get_bayes_factors yet

[PierreBoyeau]

I renamed eps to pseudocounts, because I think that the eps function for BF computation is different from the one would use as pseudocounts.
In the latter case, the objective is to zero out gene expression differences that come from softmax artifacts.