Member Junction
    Preparing search index...

    How a model's stored numbers were measured; training writes it into MLModel.Lineage under evaluation.

    interface EvaluationRecord {
        ImportanceKind:
            | "none"
            | "coefficient"
            | "split_or_impurity"
            | "held_out_permutation"
            | null;
        MajorityBaselineRule: MajorityBaselineRule;
        TimeIndexColumn: string
        | null;
        ValidationOutOfSample: boolean;
        WithheldReason: string | null;
    }
    Index

    Properties

    ImportanceKind:
        | "none"
        | "coefficient"
        | "split_or_impurity"
        | "held_out_permutation"
        | null

    Which kind of number the stored feature importance is; null for models trained before it was recorded.

    MajorityBaselineRule: MajorityBaselineRule

    The majority-baseline rule in force when the model was trained (D2-C5d).

    TimeIndexColumn: string | null

    The pipeline's declared time index (ValidationStrategy.TimeIndexColumn), or null.

    ValidationOutOfSample: boolean

    True when Metrics were measured on rows the fitted model did not train on (every strategy that writes this record).

    WithheldReason: string | null

    Why the trust verdict is withheld, or null when it is not.