BanditRLlib
Lean gate passed before this site build; local proof declarations are shown as compiled.

Lean module · EXP3

BanditRLProof.Exp3BernsteinRealizedHighProbabilityRegret

# Generated realized EXP3 regret with Bernstein predictable confidence This module composes the generated predictable Bernstein-radius EXP3 theorem with the one-sided realized-minus-exploration deviation tail. The two importance-weighted confidence events use the variance-sensitive fixed-tilt route; the realized-deviation event retains its bounded-loss Hoeffding/Azuma radius. The deterministic Hedge-square contribution also remains unchanged. Thus the endpoint controls generated selected scalar loss, but it is not a general Freedman theorem or an ideal tuned EXP3.P rate.

Module map

Declarations
3
Placeholders
0

Imports

BanditRLProof.Exp3BernsteinHighProbabilityRegret, BanditRLProof.Exp3RealizedConfidence

Imported by

BanditRLProof, BanditRLProof.Exp3BernsteinTuning

Declarations

Open an item to read its exact compact statement and source link. Detailed teaching notes are linked when registered.

def BanditRLProof.Exp3.sampledPredictableBernsteinRealizedHighProbabilityRegretBudget Compiled

Realized selected-loss regret budget whose predictable component uses the two variance-sensitive Bernstein confidence radii.

noncomputable def sampledPredictableBernsteinRealizedHighProbabilityRegretBudget {Action : Type v} (arms : Finset Action) (eta gamma : Real) (horizon : Nat) (delta : Real) : Real
theorem BanditRLProof.Exp3.sampledPredictable_bernsteinRealizedHighProbabilityRegret_tail_delta Compiled

Raw three-event form. The predictable component contributes the pure-cross and fixed-comparator Bernstein events, while the third event is the bounded realized-minus-predictable deviation.

theorem sampledPredictable_bernsteinRealizedHighProbabilityRegret_tail_delta {Env : Type u} {Action : Type v} [MeasurableSpace Env] [StandardBorelSpace Env] [Nonempty Env] [MeasurableSpace Action] [MeasurableSingletonClass Action] [StandardBorelSpace Action] [Nonempty Action] [DecidableEq Action] (prior : Measure Env) [IsProbabilityMeasure prior] (arms : Finset Action) (harms : arms.Nonempty) (eta gamma : Real) (heta : 0 < eta) (hgamma_pos : 0 < gamma) (hgamma_lt_one : gamma < 1) (loss : PredictableLossVector Env Action) (comparator : Action) (hcomparator : comparator ∈ arms) (horizon : Nat) (hhorizon : 0 < horizon) (delta : Real) (hdelta : 0 < delta) : let mu := prior ⊗ₘ sampledImportanceWeightedTrajectoryKernel arms harms eta gamma hgamma_pos.le hgamma_lt_one.le loss.environment mu {sample | sampledPredictableBernsteinRealizedHighProbabilityRegretBudget arms eta gamma horizon delta <= (Finset.range horizon).sum (fun t => sampledTrajectoryRealizedLossAt t sample) - (Finset.range horizon).sum (fun t => predictableLossAt loss t sample comparator)} <= (ENNReal.ofReal delta + ENNReal.ofReal delta) + ENNReal.ofReal delta
theorem BanditRLProof.Exp3.sampledPredictable_bernsteinRealizedHighProbabilityRegret_tail_total_delta Compiled

Total-failure form: the pure-cross Bernstein, fixed-comparator Bernstein, and realized-deviation events each receive `delta / 3`.

theorem sampledPredictable_bernsteinRealizedHighProbabilityRegret_tail_total_delta {Env : Type u} {Action : Type v} [MeasurableSpace Env] [StandardBorelSpace Env] [Nonempty Env] [MeasurableSpace Action] [MeasurableSingletonClass Action] [StandardBorelSpace Action] [Nonempty Action] [DecidableEq Action] (prior : Measure Env) [IsProbabilityMeasure prior] (arms : Finset Action) (harms : arms.Nonempty) (eta gamma : Real) (heta : 0 < eta) (hgamma_pos : 0 < gamma) (hgamma_lt_one : gamma < 1) (loss : PredictableLossVector Env Action) (comparator : Action) (hcomparator : comparator ∈ arms) (horizon : Nat) (hhorizon : 0 < horizon) (delta : Real) (hdelta : 0 < delta) : let mu := prior ⊗ₘ sampledImportanceWeightedTrajectoryKernel arms harms eta gamma hgamma_pos.le hgamma_lt_one.le loss.environment mu {sample | sampledPredictableBernsteinRealizedHighProbabilityRegretBudget arms eta gamma horizon (delta / 3) <= (Finset.range horizon).sum (fun t => sampledTrajectoryRealizedLossAt t sample) - (Finset.range horizon).sum (fun t => predictableLossAt loss t sample comparator)} <= ENNReal.ofReal delta