VeloVI different results for Eythroid_others.ipynb

[emix-c]

Hi again,
I wanted to submit another issue since I have a question pertaining to a different notebook. 'Eythroid_others.ipynb' .
I'm getting very different results when performing the trajectory inference using VeloVI on Erythroid data.
I've been following it step by step and the inference worked as expected via Slingshot and DPT.
This is what I'm seeing:

In contrast, in the github, the cells result in this output:

I didn't do anything additional to adata and the code is exactly the same, so am wondering why there is such a huge discrepancy on the number of genes being filtered out.
This is causing a very different trajectory model and very different RMSE and velocity map.

I'm using scVelo version: 0.3.3 and scanpy version: 1.10.4.

Have you encountered this issue before? Please advise!

[meichenfang]

Hi! I was using scVelo version 0.3.2, scanpy version 1.8.2, veloVI version 0.2.0.

I haven't encountered this and suspected it might be due to scv version difference. Does removing more genes improve the accuracy?