search

openproblems-bio / neurips2021_multimodal_viash

MIT License
18 stars 7 forks source link

totalVI baseline #30

Open adamgayoso opened 3 years ago

adamgayoso commented 3 years ago

Hello,

I came across this file

https://github.com/openproblems-bio/neurips2021_multimodal_viash/blob/main/src/joint_embedding/methods/baseline_totalvi/script.py

It seems as though totalVI is not using the count data (relative to how the aws hosted anndata is formatted, with count layers).

Also what does the max_epochs=20 mean at the top of the file? That would probably be far too few for a baseline.

Happy to help iterate on this baseline. Thanks!

LuckyMD commented 3 years ago

@xlancelottx @rcannood can you guys comment on this? I have been using scVI with far more epochs previously as well... but I guess this might have been done to fit within the 10 min/1 hour/longer evolving time limit that was placed?

@adamgayoso We reformatted the objects to store different data in adata.X in between. @rcannood will be able to comment more on whether this was taken into account for totalVI also in the pilot run.