dataeval.scope.Coverage.Config¶

class dataeval.scope.Coverage.Config¶

Configuration for the Coverage evaluator.

extractor¶

Feature extractor for computing embeddings.

Type:¶

FeatureExtractor or None

batch_size¶

Batch size for embedding computation.

Type:¶

int or None

method¶

Coverage radius method.

Type:¶

{“naive”, “adaptive”}, default “adaptive”

num_observations¶

Neighbors required for a sample to be covered.

Type:¶

int, default 20

percent¶

Fraction of samples to flag as uncovered (adaptive method).

Type:¶

float, default 0.01

min_class_samples¶

Minimum samples for a class’s per-class signals to be assessed.

Type:¶

int, default 20

isotropy_min_samples¶

Minimum samples for a class’s isotropy to be reported (None: dims + 1).

Type:¶

int or None, default None

near_duplicate_factor¶

Fraction of the class median nearest-neighbor distance below which a pair is a near-duplicate.

Type:¶

float, default 0.5