Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Result 3, p. 352 — population LogitBoost (two classes) uses Newton steps for the additive symmetric logistic model by maximum likelihood

Proved
AddLogReg.LogitBoost.result_3

by mikedeng1 · Oct 7, 2026 · Mathlib 0df444a (Lean v4.33.1)

boostinglogistic-regressionlogitboostnewton-methodp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1

Result 3. The LogitBoost algorithm (two classes, population version) uses Newton steps for fitting an additive symmetric logistic model by maximum likelihood.

Precisely, as made explicit by the Derivation: let ν\nuν be a probability measure on X×{0,1}X\times\{0,1\}X×{0,1}, the joint law of the feature xxx and the 0/1 response y∗y^*y∗; let F:X→RF : X\to\mathbb RF:X→R be the current additive fit and p(u)=eu/(eu+e−u)p(u) = e^{u}/(e^{u}+e^{-u})p(u)=eu/(eu+e−u) the symmetric logistic (30), with p(x)=p(F(x))p(x) = p(F(x))p(x)=p(F(x)). Fix x∈Xx\in Xx∈X and let

ℓx(t)=E[ 2y∗(F(x)+t)−log⁡(1+e2(F(x)+t)) ∣ x]\ell_x(t) = E\Big[\,2y^*(F(x)+t) - \log\big(1+e^{2(F(x)+t)}\big)\ \Big|\ x\Big]ℓx​(t)=E[2y∗(F(x)+t)−log(1+e2(F(x)+t)) ​ x]

be the conditional expected log-likelihood (31) of the update F(x)+tF(x)+tF(x)+t. Then

  1. ℓx\ell_xℓx​ has derivative s(x)=2E(y∗−p(x)∣x)s(x) = 2E(y^*-p(x)\mid x)s(x)=2E(y∗−p(x)∣x) at t=0t=0t=0;
  2. for every ttt, ℓx\ell_xℓx​ has derivative 2E(y∗−p(F(x)+t)∣x)2E(y^*-p(F(x)+t)\mid x)2E(y∗−p(F(x)+t)∣x) at ttt, and this derivative function has derivative H(x)=−4E(p(x)(1−p(x))∣x)H(x) = -4E(p(x)(1-p(x))\mid x)H(x)=−4E(p(x)(1−p(x))∣x) at t=0t = 0t=0;
  3. H(x)≠0H(x)\neq 0H(x)=0;
  4. the population LogitBoost update of Algorithm 3 — F(x)+12Ew(z∣x)F(x)+\frac12E_w(z\mid x)F(x)+21​Ew​(z∣x) with working response z=(y∗−p(x))/(p(x)(1−p(x)))z = (y^*-p(x))/(p(x)(1-p(x)))z=(y∗−p(x))/(p(x)(1−p(x))) and weight w(x)=p(x)(1−p(x))w(x) = p(x)(1-p(x))w(x)=p(x)(1−p(x)) — equals the Newton update
F(x)−H(x)−1s(x).F(x) - H(x)^{-1}s(x).F(x)−H(x)−1s(x).

This identifies LogitBoost as stagewise Newton–Raphson on the expected Bernoulli log-likelihood of the additive symmetric logistic model, the population counterpart of the iteratively reweighted least squares fit of logistic regression.

Formalization Note Labels are Bool with y∗(true)=1y^*(\texttt{true}) = 1y∗(true)=1, y∗(false)=0y^*(\texttt{false}) = 0y∗(false)=0; conditional expectations are integrals against Mathlib's regular conditional distribution ν.condKernel x, and the statement holds for every xxx. Derivatives are stated with HasDerivAt; the second derivative is the derivative of an explicit function shown to be the first derivative everywhere. H(x)≠0H(x)\ne0H(x)=0 is proved, not assumed. In the population version, step 2(b) (the weighted least-squares regression of zzz on xxx) is the weighted conditional mean Ew(z∣x)E_w(z\mid x)Ew​(z∣x), as the paragraph after (36) explains.

Preamble
import Mathlib
import Definitions.Def_AddLogReg_LogitBoost_Setting

open MeasureTheory ProbabilityTheory
Formal statement
namespace AddLogReg.LogitBoost

theorem result_3 {X : Type*} [MeasurableSpace X] (ν : Measure (X × Bool))
    [IsProbabilityMeasure ν] (F : X → ℝ) (x : X) :
    HasDerivAt (condLogLik ν F x)
        (2 * ∫ b, (ystar b - symLogistic (F x)) ∂(ν.condKernel x)) 0 ∧
      (∀ t : ℝ, HasDerivAt (condLogLik ν F x)
        (2 * ∫ b, (ystar b - symLogistic (F x + t)) ∂(ν.condKernel x)) t) ∧
      HasDerivAt (fun t : ℝ => 2 * ∫ b, (ystar b - symLogistic (F x + t)) ∂(ν.condKernel x))
        (-4 * ∫ _b, symLogistic (F x) * (1 - symLogistic (F x)) ∂(ν.condKernel x)) 0 ∧
      (-4 * ∫ _b, symLogistic (F x) * (1 - symLogistic (F x)) ∂(ν.condKernel x) ≠ 0) ∧
      logitBoostStep ν F x =
        F x - (2 * ∫ b, (ystar b - symLogistic (F x)) ∂(ν.condKernel x)) /
          (-4 * ∫ _b, symLogistic (F x) * (1 - symLogistic (F x)) ∂(ν.condKernel x)) := by sorry

end AddLogReg.LogitBoost
Source
Friedman, Hastie and Tibshirani, Additive logistic regression: a statistical view of boosting, Ann. Statist. 28 (2000), p. 352, Result 3 and its Derivation (31)–(36); p. 351, (30) and Algorithm 3
Human review
  • Endorsed by Shuze Chen · Oct 7, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 7, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me