Expand description
Outer-loop row subsampling primitive shared across the solver and the family-specific outer-score evaluators.
OuterScoreSubsample and its per-row WeightedOuterRow are the
Horvitz–Thompson row subsample consumed on outer-loop hot paths. They live
in the solver layer (below families) so that both the solver’s row-measure
machinery and the family outer-score builders can depend on them downward,
without solver reaching up into families. The stratified builders that
construct these (build_outer_score_subsample, auto_outer_score_subsample)
remain in families::marginal_slope_shared, since they depend on
family-specific fit options; they import this type downward.
Structs§
- Outer
Score Subsample - Stratified row index subsample shared across outer-loop evaluations.
- Weighted
Outer Row
Enums§
- RowSet
- Row selection for an outer-loop evaluation: either the full data (
All) or a Horvitz–ThompsonWeightedOuterRowsubsample.
Constants§
- ARROW_
ROW_ CHUNK - Deterministic row-block tiling constant for the parallel reduction paths.
Functions§
- arrow_
row_ chunk_ count - Number of
ARROW_ROW_CHUNK-sized tiles coveringn_rows.