univi.evaluation.encode_adata

Contents

univi.evaluation.encode_adata#

univi.evaluation.encode_adata(model, adata, modality, device='cpu', layer=None, X_key='X', batch_size=1024, latent='moe_mean', random_state=0)[source]#

Encode a single modality into z.

latent:

  • “moe_mean” / “moe_sample”: fused MoE/PoE posterior using only provided experts

  • “modality_mean” / “modality_sample”: that modality posterior only

Contents