incerto.active.compute_diversity_penalty

incerto.active.compute_diversity_penalty(selected, features, method='min_distance')

Compute diversity penalty for selected samples.

Parameters:

• selected (Tensor) – Indices of selected samples

• features (Tensor) – Feature representations of all samples

• method (str) – Diversity measure (‘min_distance’, ‘mean_distance’, ‘determinant’)

Return type:

Tensor

Returns:

Diversity score. For ‘min_distance’ and ‘mean_distance’, higher values indicate more diversity. For ‘determinant’, lower (more negative) values indicate more diversity.