Attribute Error while training simplescvi on tutorial lung data. Differences to full scvi?

Great package and impressive work!

I want to dive into some simple method development based on the scvi integration and found this simple-scvi code to start from: GitHub - scverse/simple-scvi: Example repo of using scvi-tools building blocks externally with simplified scVI implementation

Just to get it running and get a feeling for it I wanted to run it on the lung data used in the atlas-level integration tutorial (Atlas-level integration of lung data — scvi-tools).

But following the tutorial, just not with the normal scvi but this simple scvi model, I get the following Error while running model.train:
AttributeError: 'NoneType' object has no attribute 'numel'

Any Idea what the error is about ? And what are the main differences between the full version of scvi and this simple scvi ?

Thanks a lot!

In terms of model architecture, both models are identical I believe. The main difference is that the simpler version does not include additional analysis functions such as differential expression, etc. The code hasn’t been tested/updated in a while, so it’s likely that there’s some bug there.

In terms of the user API, both models are identical, so I would recommend going for the normal scVI if your intention is to get a feel for how it works. Nevertheless, I’ll take a look at the repo and fix the bug if possible.

thanks a lot for your super quick response and help!
As I want to build my own model at some point, I thought the simple scvi would be an easier model to start with, but I will also have a look on the source code of the normal scvi then!
would be awesome if you would manage to fix the bug.
Thank you!!