univi.evaluation.evaluate_cross_reconstruction

univi.evaluation.evaluate_cross_reconstruction#

univi.evaluation.evaluate_cross_reconstruction(model, adata_src, adata_tgt, *, src_mod, tgt_mod, device='cpu', src_layer=None, src_X_key='X', tgt_layer=None, tgt_X_key='X', batch_size=512, kind='continuous', feature_names=None, use_moe=True, decoder_prefer='auto', decoder_link=None)[source]#

README-style one-call cross-reconstruction evaluation:

  1. X_pred = cross_modal_predict(model, adata_src, src_mod -> tgt_mod)

  2. compare X_pred against adata_tgt truth (layer/X_key)

  3. return dict with:

    • “summary”

    • “per_feature”

    • “X_true”, “X_pred”, “feature_names” (for plotting)

Contents