Member Junction
    Preparing search index...

    Interface ModelTrainingLineage

    The provenance blob training writes to MJ: ML Models.Lineage (§4.3). Self-contained enough for the scoring path to re-assemble features when the pipeline row is not loaded.

    interface ModelTrainingLineage {
        asOfStrategy: AsOfStrategy;
        assembledAt: string;
        evaluation: EvaluationRecord;
        featureSteps: FeatureStepGraph;
        lockedHoldoutClassCounts: Record<string, number>;
        lockedHoldoutRowCount: number;
        outcomeConfig: OutcomeConfig;
        pipelineId: string;
        pipelineVersion: number;
        sidecarVersion: string;
        sourceBindings: SourceBinding[];
        targetEntityName: string;
        trainingRowCount: number;
    }
    Index

    Properties

    asOfStrategy assembledAt evaluation featureSteps lockedHoldoutClassCounts lockedHoldoutRowCount outcomeConfig pipelineId pipelineVersion sidecarVersion sourceBindings targetEntityName trainingRowCount

    Properties

    asOfStrategy: AsOfStrategy
    assembledAt: string

    ISO timestamp the matrix was assembled.

    evaluation: EvaluationRecord

    How the stored metrics were measured (EvaluationLineage.Key).

    featureSteps: FeatureStepGraph
    lockedHoldoutClassCounts: Record<string, number>

    Locked-holdout rows per label (label as text → row count) for a classification model; {} for regression. Lets a reader compute the holdout's majority-class rate without re-reading the rows.

    lockedHoldoutRowCount: number

    Rows in the locked holdout scored once into HoldoutMetrics.

    outcomeConfig: OutcomeConfig
    pipelineId: string
    pipelineVersion: number
    sidecarVersion: string
    sourceBindings: SourceBinding[]
    targetEntityName: string
    trainingRowCount: number

    Rows sent to the sidecar for training (holdout excluded).

    Settings

    Member Visibility

    On This Page

    Properties