All missing perturbation embeddings are set to zeros

[HarveyYan]

Thanks for sharing the codebase.

cell-load/src/cell_load/data_modules/perturbation_dataloader.py

Lines 427 to 432 in 59ce6e9

	for pert in missing:
	featurization_dict[pert] = torch.zeros(feature_dim)
	logger.info(
	"Set %d missing perturbations to zero vectors.", len(missing)
	)

I may have missed something, but given all the missing perturbations are set to 0, how would the model make perturbation-specific predictions?