sticks sum to one almost surely
ProvedNestedCRP.gem_sticks_sum_onebayesian-nonparametricsp2o-gran-per-chapterp2o-plan-paperp2o-v1probabilitytopic-models
Let be i.i.d. with and , and let . Then, almost surely,
The paper attributes this to Sethuraman (1994). It says the level distribution of a document is a genuine distribution on the levels, so that the posterior on level allocations has the right total mass.
Differences from the source:
- encoding: levels and sticks are 1-based → levels and sticks are 0-based; reason: Lean indexing; level ℓ here is the paper's level ℓ + 1, and no instance changes
- encoding: V_i i.i.d. Beta(α1, α2) with α1, α2 > 0 (Sethuraman, p.7:7) → V_i i.i.d. Beta(mπ, (1 − m)π) with m ∈ (0,1), π > 0; reason: a bijective reparametrization (m = α1/(α1+α2), π = α1+α2); the same set of laws
Preamble
import Mathlib import Definitions.Def_NestedCRP_Hyper import Definitions.Def_NestedCRP_gemLaw open MeasureTheory
Formal statement
namespace NestedCRP
/-- The sticks of GEM(m, π) sum to one almost surely. Blei–Griffiths–Jordan, *The nested Chinese restaurant process and Bayesian nonparametric
inference of topic hierarchies*, J. ACM 57(2), Article 7 (2010),
§2.3, p.7:7 (attributed to Sethuraman 1994) and p.7:8. -/
theorem gem_sticks_sum_one (H : Hyper) :
∀ᵐ U ∂gemLaw H.m H.pi, HasSum (stick U) 1 := 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, §2.3, p.7:7 and p.7:8
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.