Dirichlet moments as a ratio of normalizing constants
ProvedNestedCRP.dirichlet_momentbayesian-nonparametricsp2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1probabilitytopic-models
For a nonempty finite index set, parameters , exponents and ,
Integrating the multinomial parameters out of a node gives this ratio of Dirichlet normalizers, which is the source of the Gamma factors in Eq. (4).
Differences from the source:
- generalization: the symmetric prior with parameter η, and a ratio of normalizing constants stated in words → general parameters α, with the ratio written out; reason: the source's case is α_i = η
Preamble
import Mathlib import Definitions.Def_NestedCRP_dirichletMeasure
Formal statement
namespace NestedCRP
/-- The moments of the Dirichlet law, a ratio of its normalizing constants:
`E ∏_v U_v^{n_v} = Γ(∑ α_v) / ∏_v Γ(α_v) · ∏_v Γ(n_v + α_v) / Γ(∑ n_v + ∑ α_v)`.
Blei–Griffiths–Jordan, *The nested Chinese restaurant process and Bayesian nonparametric
inference of topic hierarchies*, J. ACM 57(2), Article 7 (2010), §5.2, p.7:17.
**Formalization Note** Stated for general `α`; the paper's topic prior is `α_v = η`. -/
theorem dirichlet_moment {ι : Type*} [Fintype ι] [Nonempty ι] (α : ι → ℝ) (hα : ∀ i, 0 < α i)
(n : ι → ℕ) :
∫ U, ∏ i, U i ^ n i ∂dirichletMeasure α =
Real.Gamma (∑ i, α i) / (∏ i, Real.Gamma (α i)) *
((∏ i, Real.Gamma ((n i : ℝ) + α i)) / Real.Gamma ((∑ i, (n i : ℝ)) + ∑ i, α i)) := by sorry
end NestedCRP
Source
Blei–Griffiths–Jordan, The nested Chinese restaurant process and Bayesian nonparametric inference of topic hierarchies, J. ACM 57(2), Article 7 (2010), https://doi.org/10.1145/1667053.1667056, §5.2, p.7:17
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.