Proposition 12, p. 613 — the swapped nearest neighbor weights satisfy Σᵢ Uₙᵢ(X) ≤ β(d, a/b)
ProvedStoneRegression.NearestNeighbor.proposition_12Fix , a coefficient row with , and nonnegative scales satisfying condition (7) with constants . Let be the nearest neighbor probability weight function (8), and let
be the weight of slot when is the query point and takes its place in the sample. Then for all points ,
where is the least number of sets in covering .
The bound says that a single point cannot carry much total weight as a neighbor of the other sample points, uniformly in and in the configuration. Through the exchangeability of the sample it yields condition (1) for nearest neighbor weights (Proposition 11).
Formalization Note The statement is deterministic: the points are fixed, and only nonnegativity of the scales and (7) (read for all points) are assumed, which are the only properties the page's proof uses. The swap acts on both the weights and the scales: the query is and the sample is with in slot . is an infimum over , and the existence of a finite cover is not assumed.
import Mathlib import Definitions.Def_StoneRegression_Criterion_Setting import Definitions.Def_StoneRegression_NearestNeighbor_Weights
namespace StoneRegression.NearestNeighbor
open MeasureTheory ProbabilityTheory Filter Topology
open scoped ENNReal NNReal
theorem proposition_12 {d : ℕ} (s : ScaleSeq d) (a b : ℝ) (ha : 0 < a) (hab : a ≤ b)
(hs0 : ∀ n x xs j, 0 ≤ s n x xs j) (h7 : Cond7 s a b)
(c : ℕ → ℕ → ℝ) (n : ℕ) (hn : 1 ≤ n) (hc : IsCoeffRow c n)
(x : EuclideanSpace ℝ (Fin d)) (xs : Fin n → EuclideanSpace ℝ (Fin d)) :
(∑ i, swapW (nnWeights c s) n x xs i) ≤ (beta d (a / b) : ℝ) := by sorry
end StoneRegression.NearestNeighbor
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.