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

Lean module · Tsallis-FTRL

BanditRLProof.TsallisScheduledConditionalMeanGap

# Conditional-mean producer for scheduled half-Tsallis expected gaps This module exposes the sampling-before-action sigma-algebra at each scheduled time. The scheduled action probability is measurable with respect to that sigma-algebra, so Mathlib's conditional-expectation pull-out theorem turns a constant conditional loss-gap law into `HasScheduledExpectedGapLaw`.

Module map

Declarations
5
Placeholders
0

Imports

BanditRLProof.TsallisScheduledExpectedGapSelfBounding

Imported by

BanditRLProof, BanditRLProof.TsallisScheduledIndependentMeanGap, BanditRLProof.TsallisScheduledTimeVaryingExpectedGap

Declarations

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

def BanditRLProof.Tsallis.sampledScheduledHalfTsallisPastSigma Compiled

Information in the generated action/reward trace strictly before the scheduled action at time `t`. At time zero there is no trace information; at time `n + 1` this is the sigma-algebra generated by the prefix through `n`.

@[reducible] def sampledScheduledHalfTsallisPastSigma {Env : Type u} {Action : Type v} [MeasurableSpace Action] (t : Nat) : MeasurableSpace (Env × ((k : Nat) -> Action × Real))
theorem BanditRLProof.Tsallis.sampledScheduledHalfTsallisPastSigma_le Compiled

The scheduled past sigma-algebra is a sub-sigma-algebra of the ambient trajectory sigma-algebra.

theorem sampledScheduledHalfTsallisPastSigma_le {Env : Type u} {Action : Type v} [MeasurableSpace Env] [MeasurableSpace Action] (t : Nat) : sampledScheduledHalfTsallisPastSigma (Env
theorem BanditRLProof.Tsallis.measurable_sampledScheduledHalfTsallisProbabilityAtTime_pastSigma Compiled

Every scheduled action-probability coordinate is measurable using only the trace prefix available before that action is sampled.

theorem measurable_sampledScheduledHalfTsallisProbabilityAtTime_pastSigma {Env : Type u} {Action : Type v} [MeasurableSpace Env] [MeasurableSpace Action] [MeasurableSingletonClass Action] [DecidableEq Action] (arms : Finset Action) (harms : arms.Nonempty) (eta : Nat -> Real) (t : Nat) (candidate : Action) (hcandidate : candidate ∈ arms) : @Measurable (Env × ((k : Nat) -> Action × Real)) Real (sampledScheduledHalfTsallisPastSigma (Env
def BanditRLProof.Tsallis.HasScheduledConditionalMeanGapLaw Compiled

Coordinatewise conditional-mean gap law relative to the information available before the scheduled action at each time.

def HasScheduledConditionalMeanGapLaw {Env : Type u} {Action : Type v} [MeasurableSpace Env] [MeasurableSpace Action] [DecidableEq Action] (mu : Measure (Env × ((k : Nat) -> Action × Real))) (arms : Finset Action) (loss : Exp3.PredictableLossVector Env Action) (best : Action) (gap : Action -> Real) (horizon : Nat) : Prop
theorem BanditRLProof.Tsallis.hasScheduledExpectedGapLaw_of_conditionalMeanGapLaw Compiled

A constant conditional loss-gap law implies the expected-gap law consumed by scheduled stochastic self-bounding. The proof uses only Mathlib's pull-out property and preservation of the integral by conditional expectation.

theorem hasScheduledExpectedGapLaw_of_conditionalMeanGapLaw {Env : Type u} {Action : Type v} [MeasurableSpace Env] [MeasurableSpace Action] [MeasurableSingletonClass Action] [DecidableEq Action] (mu : Measure (Env × ((k : Nat) -> Action × Real))) [IsProbabilityMeasure mu] (arms : Finset Action) (harms : arms.Nonempty) (eta : Nat -> Real) (loss : Exp3.PredictableLossVector Env Action) (best : Action) (gap : Action -> Real) (horizon : Nat) (hgap : HasScheduledConditionalMeanGapLaw mu arms loss best gap horizon) : HasScheduledExpectedGapLaw mu arms harms eta loss best gap horizon