§4.1, p. 345, proof of Lemma 1 — ∂E(e^{−yF(x)}|x)/∂F(x) = −P(y=1|x)e^{−F(x)} + P(y=−1|x)e^{F(x)}
ProvedAddLogReg.ExpCrit.cond_criterion_hasDerivAtboostingderivativeexponential-lossp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1
Let be the joint law of on and the conditional class probabilities. Fix and consider the conditional exponential criterion as a function of the value , . It is differentiable at every , with derivative
This is the second display of the proof of Lemma 1; Lemma 1 follows by setting this derivative to zero.
Formalization Note The derivative is stated with HasDerivAt, which asserts differentiability as well as the value, rather than with deriv, which returns at non-differentiable points.
Preamble
import Mathlib import Definitions.Def_AddLogReg_ExpCrit_Setting open MeasureTheory ProbabilityTheory
Formal statement
namespace AddLogReg.ExpCrit
/-- Proof of Lemma 1, second display (p. 345): `∂E(e^{−yF(x)}|x)/∂F(x) =
−P(y = 1|x)e^{−F(x)} + P(y = −1|x)e^{F(x)}`. -/
theorem cond_criterion_hasDerivAt {X : Type*} [MeasurableSpace X] (ν : Measure (X × Bool))
[IsProbabilityMeasure ν] (x : X) (t : ℝ) :
HasDerivAt (condCrit ν x)
(-(condProb ν true x) * Real.exp (-t) + condProb ν false x * Real.exp t) t := by sorry
end AddLogReg.ExpCrit
Source
Friedman, Hastie and Tibshirani, Additive logistic regression: a statistical view of boosting, Ann. Statist. 28 (2000), p. 345, §4.1, proof of Lemma 1, second display
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.