Output of `predict`

Report

When split_covariates contains values not in perturbation_covariates or sample_covariates, e.g. zebrafish germ layers, the predict function only returns predictions for "one germ layer source".

Now the questions is what we expect it to return. I see two options:

1. a dictionary where the condition includes the split_covariates categories, e.g. pert1_pert2_18_mesoderm
2. we take the union of the germ layers and concatenate the result internally to only contain perturbation/sample covariates, e.g. pert1_pert2_18

In any case, I have to adapt the backend to make either of those work.

Version information

No response