General randomized full-information learner
DefinitionOrderedExperts_Supplement_FullInfoPolicybandit-algorithmsonline-learningp2o-gran-per-chapterp2o-plan-paperp2o-v1regret-bounds
A full-information learner chooses one current active rank from the preceding history. Only after that choice does the complete current loss vector on the experts' active recommendations enter its history.
Formalization Note The rule has no current-loss argument, which enforces the paper's timing. Repeated recommendations simply repeat one active loss coordinate.
Definition code
import Definitions.Def_OrderedExperts_chainSize
open Finset
namespace OrderedExperts.Supplement
/-- **A general randomized full-information learner.**
Zuo, *Online Learning with Ordered Experts (via Two-point Feedback)*, STOC 2027 submission
(held version), Sec. 6, pp. 13--14.
The learner selects a current active rank from the preceding history. Only after that selection is
the current vector of losses at all expert recommendations appended to the history.
**Formalization Note.** The rule has no current-loss argument, so full information arrives after
the action as required. Repeated recommendations merely repeat one active loss coordinate. -/
structure FullInfoPolicy (A : Type*) (n : ℕ) [LinearOrder A] where
mass : (Fin n → A) → List ((Fin n → A) × ℕ × (Fin n → ℝ)) → ℕ → ℝ
nonneg : ∀ current h, ∀ a ∈ Icc 1 (chainSize (fun _ ↦ current) 0),
0 ≤ mass current h a
total : ∀ current h,
∑ a ∈ Icc 1 (chainSize (fun _ ↦ current) 0), mass current h a = 1
end OrderedExperts.Supplement
Source
Zuo, Online Learning with Ordered Experts (via Two-point Feedback), STOC 2027 submission (held version), Sec. 6, pp. 13–14