Skip to content

DiversityMetric

DiversityMetric

Bases: StrEnum

Enum for different diversity metrics.

Members

- MIN_SEPARATION:             Minimum separation of all selected items
- MEAN_SEPARATION:            Arithmetic mean separation of all selected items
- GEOMEAN_SEPARATION:         Geometric mean separation of all selected items
- APPROX_GEOMEAN_SEPARATION:  Approximate geometric mean separation of all selected items
                                  (uses faster, but still smooth approximations of log(.) and exp(.))
- NON_ZERO_SEPARATION_FRAC:   Fraction of separation values that are non-zero
- MEAN_PAIRWISE_DISTANCE:     Mean distance over all pairs of selected items
                                  (the classical max-sum diversity objective)

contribution_family cached property 

contribution_family: DiversityContributionFamily

Return the diversity-contribution family whose per-point values this metric consumes.

compute 

compute(contribution_values: NDArray[float32]) -> float32

Compute diversity metric given the selected items' per-point diversity-contribution values.

Parameters

contribution_values : NDArray[np.float32] Per-point diversity-contribution values of the selected items (for separation-family metrics: each item's separation wrt the others in the selection).

Returns:

np.float32 The computed diversity score.