Theorem 3, p. 16 — for r ≥ 0 the predictor ĉ_r is feasible in (5)
ProvedDROOptimal.Predictor.theorem_3data-driven-optimizationdistributionally-robust-optimizationlarge-deviationsp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be compact, , continuous in for each , and the distributionally robust predictor (10). If , then is continuous on and
where is the empirical distribution of independent samples from ; i.e. is feasible in (5).
This is the first half of the main result (Theorem 4): the out-of-sample disappointment of decays at rate at least under every model.
Formalization Note The rate condition is the logarithm-free RateLE of the Setting module (for every , eventually the disappointment is at most ), equivalent to the displayed with .
Preamble
import Mathlib import Definitions.Def_DROOptimal_Predictor_Setting import Definitions.Def_DROOptimal_Predictor_Problem5
Formal statement
namespace DROOptimal.Predictor
/-- Theorem 3 (Feasibility of ĉ_r), p. 16: under the standing assumptions of §2 (p. 5), if r ≥ 0
then the predictor ĉ_r is feasible in (5). -/
theorem theorem_3 {n d : ℕ} (X : Set (EuclideanSpace ℝ (Fin n))) (hX : IsCompact X)
(γ : ↥X → Fin d → ℝ) (hγ : ∀ i, Continuous (fun x : ↥X => γ x i))
(r : ℝ) (hr : 0 ≤ r) :
Feasible5 γ r (drPredictor γ r) := by sorry
end DROOptimal.Predictor
Source
Van Parys, Mohajerin Esfahani & Kuhn, From Data to Decisions: Distributionally Robust Optimization is Optimal, arXiv:1704.04118v3, p. 16, Theorem 3; proof pp. 16–17
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.