dataeval.scope.CoverageOutput¶

class dataeval.scope.CoverageOutput(data, *, uncovered_indices, coverage_radius, critical_value_radii)¶

A dataset’s per-class embedding-space coverage.

The wrapped DataFrame is the per-class breakdown — one row per class, sorted with the lowest-dispersion assessable classes first (the ones most worth collecting varied data for) and unassessable classes last — with columns class, count, uncovered, uncovered_fraction, dispersion, isotropy, near_duplicate_fraction, and assessable. isotropy and near_duplicate_fraction are null for classes below the sample floors that make them meaningful. The sample-level coverage detail hangs off it as attributes.

uncovered_indices¶

Indices of individual samples sitting in under-sampled regions of the full embedding space (from coverage_adaptive() / coverage_naive()).

Type:¶

NDArray[np.intp]

coverage_radius¶

The radius threshold separating covered from uncovered samples.

Type:¶

float

critical_value_radii¶

Per-sample distance to the num_observations-th nearest neighbor — the raw density signal the uncovered set is thresholded from.

Type:¶

NDArray[np.float32]

data()¶

Return the output data as a polars DataFrame.

meta()¶

Metadata about the execution of the function or method for the Output class.

Return type:¶

ExecutionMetadata