Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 1, p. 11 — every customer choice model is a decision forest model over the purchase decision trees of depth at most N + 1

Proved
DecisionForest.Universal.theorem_1

by mikedeng1 · Oct 23, 2026 · Mathlib 0df444a (Lean v4.33.1)

decision-forestdiscrete-choicep2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1universality

Consider NNN products N={1,…,N}\mathcal N=\{1,\dots,N\}N={1,…,N} and the purchase options N+={0,1,…,N}\mathcal N^+=\{0,1,\dots,N\}N+={0,1,…,N}, where 000 is the no-purchase option. A customer choice model P(⋅∣⋅)\mathbf P(\cdot\mid\cdot)P(⋅∣⋅) gives, for every assortment S⊆NS\subseteq\mathcal NS⊆N (the empty one included), a probability distribution P(⋅∣S)\mathbf P(\cdot\mid S)P(⋅∣S) on N+\mathcal N^+N+ with P(p∣S)=0\mathbf P(p\mid S)=0P(p∣S)=0 for every product p∉Sp\notin Sp∈/S. Let FdF_dFd​ be the set of purchase decision trees that satisfy Requirements 1–3 of §3.2 and have depth at most ddd, and for a finite forest FFF with weights λ\lambdaλ let P(F,λ)(o∣S)=∑t∈Fλt I{o=A^(S,t)}\mathbf P^{(F,\lambda)}(o\mid S)=\sum_{t\in F}\lambda_t\,\mathbb I\{o=\hat A(S,t)\}P(F,λ)(o∣S)=∑t∈F​λt​I{o=A^(S,t)} be the decision forest model (3).

Theorem 1. For every customer choice model P(⋅∣⋅)\mathbf P(\cdot\mid\cdot)P(⋅∣⋅) there is a probability distribution λ\lambdaλ over FN+1F_{N+1}FN+1​ such that

P(o∣S)=P(FN+1,λ)(o∣S)\mathbf P(o\mid S)=\mathbf P^{(F_{N+1},\lambda)}(o\mid S)P(o∣S)=P(FN+1​,λ)(o∣S)

for every assortment S⊆NS\subseteq\mathcal NS⊆N and every purchase option o∈N+o\in\mathcal N^+o∈N+.

The theorem says that the decision forest model is universal: it represents every choice model, whether or not it is a random utility model or satisfies regularity, and trees of depth N+1N+1N+1 always suffice.

Formalization Note The distribution is given by a finite set F of trees contained in FN+1F_{N+1}FN+1​ (FSet N (N + 1)) and weights lam that are nonnegative on F and sum to one over F; weights outside F never enter. Products are Fin N, options Option (Fin N) with none the no-purchase option, assortments Finset (Fin N).

Differences from the source:

  • encoding: a distribution λ\lambdaλ over FN+1F_{N+1}FN+1​ → a finite set F⊆FN+1F\subseteq F_{N+1}F⊆FN+1​ with weights nonnegative on FFF summing to one over FFF; reason: FN+1F_{N+1}FN+1​ is finite (p. 8), so every distribution on it is of this form (take F=FN+1F=F_{N+1}F=FN+1​ or its support), and P(FN+1,λ)=P(F,λ)\mathbf P^{(F_{N+1},\lambda)}=\mathbf P^{(F,\lambda)}P(FN+1​,λ)=P(F,λ) when λ\lambdaλ vanishes off FFF.
  • encoding: products {1,…,N}\{1,\dots,N\}{1,…,N} → Fin N (product kkk is k - 1), options N+\mathcal N^+N+ → Option (Fin N) with none the option 000; reason: Lean convention, same instances.
  • encoding: Requirement 1 → built into the type of trees PTree N; reason: every split carries a product by construction, same trees.
Preamble
import Mathlib
import Definitions.Def_DecisionForest_Universal_Setting
Formal statement
namespace DecisionForest.Universal

theorem theorem_1 (N : ℕ) (P : Option (Fin N) → Finset (Fin N) → ℝ) (hP : IsChoiceModel P) :
    ∃ (F : Finset (PTree N)) (lam : PTree N → ℝ),
      (↑F ⊆ FSet N (N + 1)) ∧ (∀ t ∈ F, 0 ≤ lam t) ∧ ∑ t ∈ F, lam t = 1 ∧
      ∀ (S : Finset (Fin N)) (o : Option (Fin N)), P o S = forestProb F lam o S := by sorry

end DecisionForest.Universal
Source
Chen and Mišić, Decision Forest: A Nonparametric Approach to Modeling Irrational Choice, arXiv:1904.11532v3, p. 11, Theorem 1, (4)
Human review
  • Endorsed by Shuze Chen · Oct 23, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 23, 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