dataeval.types¶

Data types used in DataEval.

Attributes¶

`AlignmentRelation`	How a `Correspondence`'s source concept relates to its target concept.
`StatsMap`	A mapping of metric names to their corresponding numpy array values.

`BaseCollectionMixin`	Mixin providing collection interface for Output subclasses.
`ClusterConfigMixin`	Configuration mixin for evaluators that use clustering.
`Correspondence`	One typed mapping between a source and a target concept in an alignment.
`DatasetInfo`	Descriptive metadata for a dataset artifact.
`Evaluator`	Base class for all evaluators.
`EvaluatorConfig`	Base configuration class for all evaluators.
`ExecutionMetadata`	Metadata about the execution of the function or method for the Output class.
`ExtractorInfo`	Descriptor for a feature extractor as used to produce embeddings.
`MetadataJson`	Top-level schema for a `metadata.json` sidecar file.
`ModelInfo`	Descriptive metadata for a model artifact.
`OntologyConcept`	A single concept (class) within an ontology.
`ReprMixin`	Mixin providing consistent `__repr__` via `__init__` signature introspection.
`SelectionInfo`	Descriptor for a single `dataeval.data` selection step.
`SourceIndex`	The indices of the source item, target and channel.
`Track`	All observations of a single object track within one video sequence.
`TransformInfo`	Descriptor for a single image transform (typically a torchvision transform).

`DataFrameOutput`	An Output that wraps a Polars DataFrame and proxies its interface.
`DictOutput`	An Output that exposes its public instance attributes as a dictionary.
`MappingOutput`	An Output that wraps a mapping and proxies its interface.
`Output`	Base class for all evaluator output types.
`SequenceOutput`	An Output that wraps a sequence and proxies its interface.

`set_metadata`	Stamp Output classes with runtime metadata.

type dataeval.types.AlignmentRelation = 'equivalent' | 'narrower' | 'broader' | 'related'¶

How a Correspondence’s source concept relates to its target concept.

The vocabulary mirrors the W3C SKOS mapping properties:

"equivalent" (skos:exactMatch / owl:equivalentClass) — the two concepts denote the same class; the source label can be renamed losslessly.
"narrower" (skos:narrowMatch) — the source is more specific than the target; the source label can be safely coarsened up to the target.
"broader" (skos:broadMatch) — the source is more general than the target; carrying it over would require splitting it into finer targets, so the relation alone does not license a rewrite (diagnostic of a granularity mismatch).
"related" (skos:relatedMatch) — the two are associated (e.g. share an ancestor) but neither subsumes the other; not a label rewrite.

type dataeval.types.StatsMap = Mapping[str, NDArray[Any]]¶: A mapping of metric names to their corresponding numpy array values. Each array should have the same length along the first dimension, representing the number of samples.