Proposition 7.4: for a target process in a favourable state, the Gittins index equals the current probability of success
ProvedAllocationIndices.favourable_index_eq_cpsfavourable-priorsgittins-indextarget-processes
Proposition 7.4. If is a target process and the probability density for the parameter is favourable, then .
Formally: for a conjugate sampling model, a target , a state that is favourable for (along every finite sequence of observations below the current probability of success never exceeds its value at ) and , the Gittins index of the target process in the state equals its current probability of success .
Preamble
import Definitions.Def_AllocationIndices_Sampling open MeasureTheory ProbabilityTheory BanditAlgorithm
Formal statement
namespace AllocationIndices
/-- **Proposition 7.4** (p. 181). If `S` is a target process and the probability density `π` for
the parameter `θ` is favourable, then `ν(S, π) = r(S, π)`: the Gittins index of the target
process in a favourable state equals its current probability of success. -/
theorem favourable_index_eq_cps {Θ P : Type*} [MeasurableSpace Θ] [StandardBorelSpace Θ]
[Nonempty Θ] [MeasurableSpace P] (F : SamplingModel Θ P) (hconj : F.IsConjugate) (T : ℝ)
(p : P) (hfav : F.IsFavourable T p) {a : ℝ} (ha0 : 0 < a) (ha1 : a < 1) :
gittinsIndex (F.targetChain T) (F.targetReward T) a (Sum.inl p) =
F.targetReward T (Sum.inl p) := by sorry
end AllocationIndices
Source
Gittins, Glazebrook and Weber, Multi-armed Bandit Allocation Indices, 2nd ed., Wiley 2011, doi:10.1002/9780470980033, §7.3 p. 181, Proposition 7.4 (from Proposition 2.7)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.