univi.evaluation.evaluate_alignment

univi.evaluation.evaluate_alignment#

univi.evaluation.evaluate_alignment(*, Z1, Z2, metric='euclidean', recall_ks=(1, 5, 10), k_mixing=20, k_entropy=30, labels_source=None, labels_target=None, compute_bidirectional_transfer=True, k_transfer=15, json_safe=True, block_size=512)[source]#

README-facing alignment evaluation.

Computes:

  • FOSCTTM (mean + SEM)

  • Recall@k (mean + SEM) for each k

  • modality mixing + entropy in the stacked space

  • label transfer (source->target), and optionally bidirectional summary

Returns a flat dict with stable keys used by README examples.

Contents