Member Junction
    Preparing search index...

    An experiment session's leaderboard entry: the Core entry plus the holdout metric its normalized Metric was read from, so the leaderboard (the session's snapshot and the run's result) states each entry's metric.

    interface ExperimentLeaderboardEntry {
        IterationID: string;
        Metric: number;
        ModelID?: string;
        OnSuccessMetric: boolean;
        ScoreMetric: string;
    }

    Hierarchy (View Summary)

    Index

    Properties

    IterationID: string

    The iteration this entry scores (MJ: Experiment Session Iterations id).

    Metric: number

    The normalized leaderboard metric value (the Experiment's TargetMetric).

    ModelID?: string

    The trained model this iteration produced, if any (pruned/failed runs have none).

    OnSuccessMetric: boolean

    True when ScoreMetric is the plan's success metric. Numbers of two metrics are not comparable, so an entry on the success metric ranks above every entry on a fallback metric: a kept simple baseline is the model a shipped finalist of the same unit and protocol beat at its own gate.

    ScoreMetric: string

    The holdout metric Metric was read from: the plan's success metric, or the model's primary proper score when its holdout gave no score on the success metric (a kept simple baseline stores its primary score alone).