univi.workflows.make_loader#
- univi.workflows.make_loader(adata_by_mod, *, batch_size=256, shuffle=False, labels=None, recon_targets_spec=None, drop_last=False, seed=0)[source]#
Build a CPU dataset with the required UniVI collator.
A mapping with multiple modalities must contain genuinely paired cells. Use separate loaders for unrelated cohorts. BatchNorm training requires at least two cells per batch; use drop_last=True for a singleton remainder.