agrag.eval.resolution.cluster_quality_metric
cluster_quality_metric(*, threshold:float = 0.5) -> ScoreMetric
Build a metric for cluster quality on one test case.
The score is B-cubed F1. The breakdown holds b_cubed_precision,
b_cubed_recall, pairwise_precision, pairwise_recall and
pairwise_f. An over-merge lowers precision and an under-merge lowers
recall. The score of a whole dataset is the score of one case that holds
all its mentions, because pooling clusters from separate cases is not
defined.
Parameters:
- threshold (
float) – The minimum score that counts as success.
Returns:
ScoreMetric– A metric that scores one case fromresolution_case.