vertebrae.datasets.retrieval

Dataset contracts for exact frozen-embedding retrieval evaluation.

Classes

RetrievalDataset

Independent query and gallery collections with explicit graded relevance.

Module Contents

class vertebrae.datasets.retrieval.RetrievalDataset[source]

Independent query and gallery collections with explicit graded relevance.

classmethod from_relevance_matrix(queries, gallery, relevance, **kwargs)[source]

Alias that makes a dense relevance-matrix workflow explicit.

Parameters:

  • queries (Any)

  • gallery (Any)

  • relevance (Any)

  • kwargs (Any)

Return type:

RetrievalDataset

query_id_values()[source]

Return the validated query-ID snapshot used by the protocol.

Return type:

Tuple[Any, Ellipsis]

gallery_id_values()[source]

Return the validated gallery-ID snapshot used by the protocol.

Return type:

Tuple[Any, Ellipsis]

query_values()[source]

Return a detached copy of the validated query inputs used for evaluation.

Return type:

Any

gallery_values()[source]

Return a detached copy of the validated gallery inputs used for evaluation.

Return type:

Any

normalized_relevance()[source]

Return a copy of the validated relevance snapshot.

Return type:

Dict[int, Dict[int, float]]

normalized_exclusions()[source]

Return a copy of the validated exclusion snapshot.

Return type:

set[Tuple[int, int]]

protocol_modalities()[source]

Return the validated query and gallery modalities.

Return type:

Tuple[str, str]