Member Junction
    Preparing search index...
    interface AsOfAssemblyInput {
        AssembledAt: Date;
        CategoryLevels?: ReadonlyMap<string, readonly string[]>;
        Context: "train" | "score";
        DataAccess: ICountingDataAccess;
        DevelopmentRowsOf?: (rows: readonly AssembledRow[]) => ReadonlySet<string>;
        ExplicitOps?: readonly PreprocessingOp[];
        Features: readonly FeatureColumnSpec[];
        History: IRecordHistoryReader;
        Metadata: IAssemblyMetadata;
        Read: ReadPolicy;
        RowEntity: AssemblyEntityInfo;
        Rows: readonly AssemblyRowRef[];
        SeriesHistory: SeriesHistory;
        Unit: PredictionUnitSpec;
    }
    Index

    Properties

    AssembledAt: Date
    CategoryLevels?: ReadonlyMap<string, readonly string[]>

    The kept levels of each categorical column, by column name, as training produced them (the result's CategoryLevels). Scoring must pass training's levels so a value folds into __other exactly as it did in training; a column listed here is folded to these levels without recounting. Absent (training), the levels are computed from the assembled rows.

    Context: "train" | "score"
    DevelopmentRowsOf?: (rows: readonly AssembledRow[]) => ReadonlySet<string>

    Training only: the ids of the rows, among the assembled ones, whose values the categorical fold counts its levels over (the carve's development rows). The other rows (the locked holdout) are folded to those levels as scoring folds a row, so a level seen only there becomes __other and never enters the kept levels. Absent, every assembled row is counted.

    ExplicitOps?: readonly PreprocessingOp[]
    Features: readonly FeatureColumnSpec[]
    Rows: readonly AssemblyRowRef[]
    SeriesHistory: SeriesHistory