AFTD Auto-Formalizing Theoretical Domains

plackett_luce_surprisal_measure

def

For weights w : ℕ → ℝ and number of items n : ℕ, the pushforward measure on ℝ of the standardized surprisal (-log P(l) - H) / sqrt(n) under the Plackett-Luce distribution on rankings of Finset.range n, where H is the Plackett-Luce entropy of Finset.range n.

Statement

noncomputable def plackett_luce_surprisal_measure (w : ℕ → ℝ) (n : ℕ) : MeasureTheory.Measure ℝ
Defined in
OP-1712
Verified
14 Nov 2026
Axioms
none listed
Built on 3
  • plackett_luce_entropyShannon entropy (natural logarithm) of the Plackett-Luce distribution over the rankings of S.
  • plackett_luce_probPlackett-Luce probability of a ranking (best first): the product over positions of the weight of the alternative placed there divided by…
  • rankings_ofThe rankings of a finite set S of alternatives: all orderings of S as duplicate-free lists.

Read back from the Lean

A definition of a measure on ℝ, parameterized by a weight function w : ℕ → ℝ and a natural number n : ℕ. It defines the measure as the finite sum over all permutations l of {0, …, n - 1} of the Dirac delta measure at (-Real.log (plackett_luce_prob w l) - H) / Real.sqrt n scaled by ENNReal.ofReal (plackett_luce_prob w l), where H is the Plackett-Luce entropy of w on {0, …, n - 1}.

Written by a model that saw only the Lean, never the English above. If the two disagree, that is worth a challenge.

Lean source view module on GitHub

/-- Standardized surprisal distribution for Plackett-Luce rankings of n alternatives. -/
noncomputable def plackett_luce_surprisal_measure (w : ℕ → ℝ) (n : ℕ) : MeasureTheory.Measure ℝ := let H := plackett_luce_entropy w (Finset.range n)
  ∑ l ∈ rankings_of (Finset.range n),
    ENNReal.ofReal (plackett_luce_prob w l) • MeasureTheory.Measure.dirac ((-Real.log (plackett_luce_prob w l) - H) / Real.sqrt (n : ℝ))
See it in the graphDiscuss this resultChallenge it