The suffix appended to a metric name to reflect the aggregation strategy,
e.g. "precision" vs. "precision_micro". No suffix is used when
positiveLabel is explicitly configured, since average is not
applicable in that one-vs-rest binary mode. This mirrors
resolvePositiveLabel's auto-detection rule (only under the default
"macro" average) so a constructed evaluator's static name always
matches what computePrecisionRecallFScore actually computes.
The suffix appended to a metric name to reflect the aggregation strategy, e.g.
"precision"vs."precision_micro". No suffix is used whenpositiveLabelis explicitly configured, sinceaverageis not applicable in that one-vs-rest binary mode. This mirrorsresolvePositiveLabel's auto-detection rule (only under the default"macro"average) so a constructed evaluator's static name always matches whatcomputePrecisionRecallFScoreactually computes.