David S. Fischer

Hi @wangjiawen2013, thanks for the issue! I just resolved this, the fix is on diffxpy and batchglm dev branches so please install them from source via `pip install -e .`....

> My adata stores scaled data, and adata.raw stores raw counts, could you give me an example on how to test it using negative binomial noise models ? adata.raw only...

I will make access of raw objects easier, I will comment in this issue again once it s done! Thanks for the feedback @wangjiawen2013!

Hi @slieped! You can fit NB models also on data that does not only contain integers, the further you stray from NB distributed data, the less this analysis makes sense...

> Hey, > > In my Wald tests, lowest non-zero p-value I get is 2.220446049250313e-16. I'm using master branch and default dtype seems float64. I was wondering why it's not...

You can also start a PR onto dev just fixing this doc string!

Thanks! You could try this: https://github.com/theislab/sfaira/blob/a713ab0a679e2b409ae841f3d35bebda758c413d/sfaira/data/base/dataset.py#L28 and https://github.com/theislab/sfaira/blob/a713ab0a679e2b409ae841f3d35bebda758c413d/sfaira/data/base/dataset.py#L416

Hi @dburkhardt, thanks for the detailed issue! Somehow, two_sample escaped the unit tests so far, sorry for that. This error `ValueError: cannot model more than one scale parameter with numpy...

Note that this only makes sense for Wald test. This does not work for LRT, T-test and Wilcoxon test, which is why this is only a speed-up enhancement but not...

This would be in ./test/base.py:_Partition.wald()