pass custom pytorch dataset with custom __get_item__ method to the SCVI

[igor-sadalski]

While there are some ways to pass custom data loaders (https://docs.scvi-tools.org/en/1.3.2/user_guide/use_case/custom_dataloaders.html) these approaches seem a bit of an overkill. On the other hand needing to concatenate your data (if you have multiple datasets) gets very cumbersome very quickly.

t would be nice if you can just create a custom pytorch dataset and pass it to the SCVI api. This dataset would have the get_item_ method that just returns the needed transcriptome and say batch etc.

[ori-kron-wis]

Hey @igor-sadalski ,
Can you elaborate on your use case? Why can't it be handled with the recent version API (note 1.4v is out)?

We gave several examples of common custom dataloaders implementations use cases (tiledb, lamin, anncollection) but you can also use your wildcard dataloader in the same manner (but you will have to develop it).