About weight and training problems

Thank you for your excellent tutorial.

I read the article and code. It seems you split the raw data into source data and target data. I have a question why train VAE separately with respect to different data? Could I this using the source_datasets is trained the model, then using target_datasets trained is like fine-tuning?

And I also see there are two encoders. The front part seems similar, 1003128, 128128. Is these two parts weight shared?

Could you specify which tutorial you’re referring to?

here is the link
(Multi-Modal Surgery Pipeline with TOTALVI — scArches documentation)

For scArches usage, I recommend you make an issue on their github page (GitHub - theislab/scarches: architecture surgery and transfer learning for single-cell)