Unable to pass batch_key to AmortizedLDA model


I might be fundamentally misunderstanding something, but I’d like to perform Topic Modeling with Amortized LDA and when I run through that tutorial on my own data, the setting up the model includes a line that says No batch_key inputted, assuming all cells are same batch. I thought it might be a good idea to include a batch_key (as I have batches) but discovered I was unable to pass one as I did when setting up an scVI model.

Should I be able to provide a batch_key or is this something that isn’t appropriate for topic analysis? If it isn’t necessary, perhaps adjusting the info message to not say “assuming all cells are same batch” so that people don’t get worried they are missing something?


Hi Andrew,

Thanks for giving the LDA model a try. This is a spurious print statement which will be removed in the next version which we are actively working on. Currently the model specific setup_anndata functions all use the same (legacy) function under the hood, but the kwargs specific to each model’s setup_anndata function will tell you which fields will actually be referenced by the model. If batch_key is not present in the function signature, this means that the model does not have batch correction implemented as is the case with AmortizedLDA.