proper_score_of_exposure
def
The score of forecast x on outcome j for the proper scoring rule with expected reward G and exposure g: G(x) + <g(x), e_j - x>.
Statement
noncomputable def proper_score_of_exposure {n : ℕ} (G : (Fin n → ℝ) → ℝ) (g : (Fin n → ℝ) → Fin n → ℝ) (x : Fin n → ℝ) (j : Fin n) : ℝ
- Defined in
- arXiv:2102.07081 v2, Sec. 3 (Eq. 2 score from expected reward and exposure; Def. 3.6 and Remark 3.7 QA pool as arg min of G(x) - <x, sum_i w_i g(p_i)>) and App. D (Def. D.8 Tsallis rules, here for natural gamma); formalized by the operator (Claude), 2026-10-03
- Verified
- 03 Oct 2026
- Axioms
- none listed
- Used by 1
- tsallis_three_qa_weight_score_not_concaveWithout convex exposure the QA weight-score need not be concave: for the cubic Tsallis rule, forecasts (0,0,1), (1/4,3/4,0), outcome 1,…
Lean source view module on GitHub
noncomputable def proper_score_of_exposure {n : ℕ} (G : (Fin n → ℝ) → ℝ) (g : (Fin n → ℝ) → Fin n → ℝ) (x : Fin n → ℝ) (j : Fin n) : ℝ := G x + ∑ k, g x k * ((if k = j then 1 else 0) - x k)