bestExpertLoss
defimported Lean
The best expert loss in hindsight is the minimum cumulative loss achieved by any single expert: \[ L^* \;=\; \min_{i \in [N]} L_T(i) \;=\; \inf_{i : \mathrm{Fin}\,N} \,L_T(i). \]
Statement
set_option relaxedAutoImplicit false in set_option autoImplicit false in open Real Finset BigOperators in noncomputable def bestExpertLoss {N T : ℕ} (ℓ : LossSeq N T) : ℝ
- Defined in
- TCSlib, `bestExpertLoss`
- Lean proof
- By Karim Abdel Sadek, Mark Bedaywi, from github.com/Shilun-Allan-Li/tcslib at
c8a02591b1e5(Copyright (c) 2026 Karim Abdel Sadek and Mark Bedaywi. All rights reserved, Apache-2.0). Compiled here against leanprover/lean4:v4.33.1, Mathlib v4.33.1; written for leanprover/lean4:v4.25.0.bestExpertLoss← bestExpertLoss (TCSlib/LearningTheory/Hedge/Basic.lean:124): verbatim
- Verified
- 07 Oct 2026
- Axioms
- none listed
- Built on 2
- Used by 4
- hedge_constructionHedge yields an approximate minimax pair. Let $G$ be a finite two-player zero-sum game with $M \ge 2$ row actions and $N \ge 1$ column…
- regretThe regret of the Hedge algorithm is the excess cumulative loss over the best expert in hindsight:…
- regret_le_of_log_potential_stepRegret from a per-step log-potential bound. Let $N \ge 1$, let $\eta > 0$, let $\ell$ be a loss sequence for $N$ experts over $T$ rounds,…
- potential_ge_best_expertPotential lower bound via the best expert. Fix $N \ge 1$ experts and $T$ rounds, a loss sequence $\ell$, and a learning rate $\eta > 0$.…
Lean source view module on GitHub
set_option relaxedAutoImplicit false in set_option autoImplicit false in open Real Finset BigOperators in /-- The cumulative loss `min_i L_T(i)` of the best expert in hindsight [FS97, §2]; [CBL06, §2.1]. This is an infimum over a finite nonempty set of experts, so later we can choose an expert attaining it when lower-bounding the final potential. -/ noncomputable def bestExpertLoss {N T : ℕ} (ℓ : LossSeq N T) : ℝ := ⨅ i : Fin N, cumLoss ℓ T i