Proposition 6(a) — the mean minimizes the SPO+ risk
ProvedSmartPTO.Fisher.prop6a_mean_minimizesbayes-riskcentral-symmetryp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1spo-plus-loss
Let be an integrable probability law on cost vectors in , with mean . Let be nonempty, compact, and convex, and let be an optimal-decision oracle. Suppose is continuous on all of and centrally symmetric about . Then, for every prediction ,
The mean is therefore a population minimizer of the surrogate risk.
Formalization Note “Continuous on all” means absolute continuity and full support. The latter is retained here because it is part of the paper's common premise for both parts of Proposition 6. Risk takes values in ; integrability of costs gives the finite-mean convention of §4.1.
Preamble
import Mathlib import Definitions.Def_SPOBounds_Natarajan_Model import Definitions.Def_SmartPTO_Fisher_Setting open scoped InnerProductSpace open MeasureTheory ProbabilityTheory
Formal statement
namespace SmartPTO.Fisher
/-- Proposition 6(a), p. 23: the mean minimizes SPO+ risk. -/
theorem prop6a_mean_minimizes {d : ℕ} (S : Set (EuclideanSpace ℝ (Fin d)))
(hSne : S.Nonempty) (hScpt : IsCompact S) (hScvx : Convex ℝ S)
(wstar : EuclideanSpace ℝ (Fin d) → EuclideanSpace ℝ (Fin d))
(hw : SPOBounds.Natarajan.IsOracle S wstar)
(P : Measure (EuclideanSpace ℝ (Fin d))) [IsProbabilityMeasure P]
(hInt : Integrable id P) (hCont : ContinuousOnAll P)
(hSymm : CentrallySymmetric P)
(chat : EuclideanSpace ℝ (Fin d)) :
spoPlusRisk S wstar P (∫ c, c ∂P) ≤ spoPlusRisk S wstar P chat := by sorry
end SmartPTO.Fisher
Source
Elmachtoub & Grigas, Smart "Predict, then Optimize", arXiv:1710.08005v5, p. 23, Proposition 6(a); §4.1 p. 22
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.