1 lightning.finite_horizon.studies.development_ranking

lightning.finite_horizon.studies.development_ranking

Float32 candidate-order diagnostics reduced over independent physical scenes.

Only label-separated pairs count; prediction ties earn no correct pair. Within scene pair counts describe support, while bootstrap draws resample whole scenes. These diagnostic thresholds do not revise the frozen development-regret decision.

1.1 Functions

Name	Description
candidate_pair_ordering	Return correctly ordered and informative unordered candidate pairs.
reduce_development_ranking	Recompute descriptive H1 candidate ordering by physical scene.
summarize_scene_ordering	Give each physical scene one vote, whatever its informative pair count.

1.1.1 candidate_pair_ordering

lightning.finite_horizon.studies.development_ranking.candidate_pair_ordering(
    labels,
    predictions,
    *,
    tie_band,
)

Return correctly ordered and informative unordered candidate pairs.

A prediction tie earns no correct pair. The tie band applies to the label difference only; it is not an accuracy tolerance for predicted order.

1.1.2 reduce_development_ranking

lightning.finite_horizon.studies.development_ranking.reduce_development_ranking(
    assignment,
    evaluations,
    *,
    bootstrap_seed=2930,
    bootstrap_resamples=10000,
)

Recompute descriptive H1 candidate ordering by physical scene.

evaluations is the same four learner/constant fold pairs consumed by the canonical development-screen reader. This reducer validates that whole paired source domain first, then independently recomputes the pair counts instead of trusting the recorded pair-accuracy columns.

1.1.3 summarize_scene_ordering

lightning.finite_horizon.studies.development_ranking.summarize_scene_ordering(
    per_scene,
    *,
    bootstrap_seed=2930,
    bootstrap_resamples=10000,
)

Give each physical scene one vote, whatever its informative pair count.

The shared sampler resamples scene accuracy summaries using the declared historical settings. One scene retains its empirical degenerate interval; that sparse diagnostic does not establish population uncertainty or useful generalization. These intervals do not revise the frozen regret gate.

# lightning.finite_horizon.studies.development_ranking { #aria_nbv.lightning.finite_horizon.studies.development_ranking } `lightning.finite_horizon.studies.development_ranking` Float32 candidate-order diagnostics reduced over independent physical scenes. Only label-separated pairs count; prediction ties earn no correct pair. Within scene pair counts describe support, while bootstrap draws resample whole scenes. These diagnostic thresholds do not revise the frozen development-regret decision. ## Functions | Name | Description | | --- | --- | | [candidate_pair_ordering](#aria_nbv.lightning.finite_horizon.studies.development_ranking.candidate_pair_ordering) | Return correctly ordered and informative unordered candidate pairs. | | [reduce_development_ranking](#aria_nbv.lightning.finite_horizon.studies.development_ranking.reduce_development_ranking) | Recompute descriptive H1 candidate ordering by physical scene. | | [summarize_scene_ordering](#aria_nbv.lightning.finite_horizon.studies.development_ranking.summarize_scene_ordering) | Give each physical scene one vote, whatever its informative pair count. | ### candidate_pair_ordering { #aria_nbv.lightning.finite_horizon.studies.development_ranking.candidate_pair_ordering } ```python lightning.finite_horizon.studies.development_ranking.candidate_pair_ordering( labels, predictions, *, tie_band, ) ``` Return correctly ordered and informative unordered candidate pairs. A prediction tie earns no correct pair. The tie band applies to the label difference only; it is not an accuracy tolerance for predicted order. ### reduce_development_ranking { #aria_nbv.lightning.finite_horizon.studies.development_ranking.reduce_development_ranking } ```python lightning.finite_horizon.studies.development_ranking.reduce_development_ranking( assignment, evaluations, *, bootstrap_seed=2930, bootstrap_resamples=10000, ) ``` Recompute descriptive H1 candidate ordering by physical scene. `evaluations` is the same four learner/constant fold pairs consumed by the canonical development-screen reader. This reducer validates that whole paired source domain first, then independently recomputes the pair counts instead of trusting the recorded pair-accuracy columns. ### summarize_scene_ordering { #aria_nbv.lightning.finite_horizon.studies.development_ranking.summarize_scene_ordering } ```python lightning.finite_horizon.studies.development_ranking.summarize_scene_ordering( per_scene, *, bootstrap_seed=2930, bootstrap_resamples=10000, ) ``` Give each physical scene one vote, whatever its informative pair count. The shared sampler resamples scene accuracy summaries using the declared historical settings. One scene retains its empirical degenerate interval; that sparse diagnostic does not establish population uncertainty or useful generalization. These intervals do not revise the frozen regret gate.