Arize Phoenix TS
    Preparing search index...
    AverageType: "macro" | "micro" | "weighted"

    The strategy used to aggregate per-class precision/recall/F-score into a single number when there are more than two classes (or no positiveLabel is configured).

    • "macro": unweighted mean across classes.
    • "micro": pool true/false positives and false negatives across classes before computing the metric.
    • "weighted": mean across classes, weighted by each class's support (number of true instances).