Lean module · OFUL
BanditRLProof.OFULAllTimeConfidence
# All-time scheduled OFUL confidence This module upgrades the fixed-time scalar-ridge confidence theorem to one countable union over every deterministic horizon on the same probability space. A telescoping confidence schedule allocates the total failure budget exactly: `delta_n = delta / ((n + 1) * (n + 2))`. This is an all-time confidence theorem for one fixed process. It does not identify a generated OFUL policy; a one-policy anytime regret consumer must first use a history algorithm whose confidence parameter follows this schedule.
Module map
Imports
BanditRLProof.OFULUniformTimeConfidence
Imported by
Declarations
Open an item to read its exact compact statement and source link. Detailed teaching notes are linked when registered.
def
BanditRLProof.OFUL.allTimeScheduledScalarRidgeConfidenceFailureSet
Compiled
Countable union of scalar-ridge confidence failures over every horizon.
def allTimeScheduledScalarRidgeConfidenceFailureSet {Omega : Type u} {Feature : Type v} [Fintype Feature] [DecidableEq Feature] (lambda : Real) (thetaStar : Feature -> Real) (S : Real) (feature : Nat -> Omega -> Feature -> Real) (response : Nat -> Omega -> Real) (R : Real) (deltaAt : Nat -> Real) : Set Omega
theorem
BanditRLProof.OFUL.mem_allTimeScheduledScalarRidgeConfidenceFailureSet_iff
Compiled
Membership means failure at at least one deterministic horizon.
theorem mem_allTimeScheduledScalarRidgeConfidenceFailureSet_iff {Omega : Type u} {Feature : Type v} [Fintype Feature] [DecidableEq Feature] (lambda : Real) (thetaStar : Feature -> Real) (S : Real) (feature : Nat -> Omega -> Feature -> Real) (response : Nat -> Omega -> Real) (R : Real) (deltaAt : Nat -> Real) (omega : Omega) : omega ∈ allTimeScheduledScalarRidgeConfidenceFailureSet lambda thetaStar S feature response R deltaAt ↔ ∃ n, omega ∈ scalarRidgeConfidenceFailureAt lambda thetaStar S feature response R (deltaAt n) n
theorem
BanditRLProof.OFUL.not_mem_allTimeScheduledScalarRidgeConfidenceFailureSet_iff
Compiled
Outside the countable failure union, every deterministic-horizon confidence ellipsoid holds simultaneously.
theorem not_mem_allTimeScheduledScalarRidgeConfidenceFailureSet_iff {Omega : Type u} {Feature : Type v} [Fintype Feature] [DecidableEq Feature] (lambda : Real) (thetaStar : Feature -> Real) (S : Real) (feature : Nat -> Omega -> Feature -> Real) (response : Nat -> Omega -> Real) (R : Real) (deltaAt : Nat -> Real) (omega : Omega) : omega ∉ allTimeScheduledScalarRidgeConfidenceFailureSet lambda thetaStar S feature response R deltaAt ↔ ∀ n, matrixNorm (Matrix.scalar Feature lambda + finiteHorizonFeatureGram feature n omega) (finiteHorizonRidgeEstimate (Matrix.scalar Feature lambda) feature response n omega - thetaStar) <= finiteHorizonScalarConfidenceRadius feature R (deltaAt n) lambda S n omega
theorem
BanditRLProof.OFUL.measure_allTimeScheduledScalarRidgeConfidenceFailureSet_le_tsum
Compiled
All-time scheduled confidence: countable subadditivity bounds the failure probability by the `ENNReal` sum of the fixed-time budgets.
theorem measure_allTimeScheduledScalarRidgeConfidenceFailureSet_le_tsum {Omega : Type u} {Feature : Type v} [mOmega : MeasurableSpace Omega] [StandardBorelSpace Omega] [Fintype Feature] [DecidableEq Feature] [Nonempty Feature] (mu : Measure Omega) [IsProbabilityMeasure mu] (lambda : Real) (hlambda : 0 < lambda) (thetaStar : Feature -> Real) (S : Real) (htheta : euclideanLength thetaStar <= S) (F : Filtration Nat mOmega) (feature : Nat -> Omega -> Feature -> Real) (response noise : Nat -> Omega -> Real) (R : Real) (hR : 0 < R) (projectionBound : EuclideanSpace Real Feature -> Nat -> Real) (hfeature : forall i j, StronglyMeasurable[F i] (fun omega => feature i omega j)) (hnoise : StronglyAdapted F (fun t omega => match t with | 0 => 0 | i + 1 => noise i omega)) (hprojectionBound_nonneg : forall theta i, 0 <= projectionBound theta i) (hprojectionBound : forall theta i omega, |dotProduct (WithLp.ofLp theta) (feature i omega)| <= projectionBound theta i) (hsubGaussian : forall i, HasCondSubgaussianMGF (F i) (F.le i) (noise i) (constantSquaredVarianceProxy R i) mu) (hresponse : forall omega i, response i omega = dotProduct thetaStar (feature i omega) + noise i omega) (deltaAt : Nat -> Real) (hdeltaAt : forall n, 0 < deltaAt n) (hdeltaAt_one : forall n, deltaAt n <= 1) : mu (allTimeScheduledScalarRidgeConfidenceFailureSet lambda thetaStar S feature response R deltaAt) <= ∑' n, ENNReal.ofReal (deltaAt n)
def
BanditRLProof.OFUL.allTimeTelescopingWeight
Compiled
Positive telescoping weight `1 / ((n+1)(n+2))`.
noncomputable def allTimeTelescopingWeight (n : Nat) : Real
theorem
BanditRLProof.OFUL.allTimeTelescopingWeight_eq_sub
Compiled
The telescoping weight is a difference of consecutive reciprocals.
theorem allTimeTelescopingWeight_eq_sub (n : Nat) : allTimeTelescopingWeight n = 1 / (((n + 1 : Nat) : Real)) - 1 / (((n + 2 : Nat) : Real))
theorem
BanditRLProof.OFUL.sum_range_allTimeTelescopingWeight
Compiled
Exact finite partial sum of the telescoping weights.
theorem sum_range_allTimeTelescopingWeight (n : Nat) : (Finset.range n).sum allTimeTelescopingWeight = 1 - 1 / (((n + 1 : Nat) : Real))
theorem
BanditRLProof.OFUL.allTimeTelescopingWeight_nonneg
Compiled
Every telescoping confidence weight is nonnegative.
theorem allTimeTelescopingWeight_nonneg (n : Nat) : 0 <= allTimeTelescopingWeight n
theorem
BanditRLProof.OFUL.allTimeTelescopingWeight_le_one
Compiled
Every telescoping confidence weight is at most one.
theorem allTimeTelescopingWeight_le_one (n : Nat) : allTimeTelescopingWeight n <= 1
theorem
BanditRLProof.OFUL.hasSum_allTimeTelescopingWeight
Compiled
The telescoping weights sum exactly to one.
theorem hasSum_allTimeTelescopingWeight : HasSum allTimeTelescopingWeight 1
def
BanditRLProof.OFUL.allTimeTelescopingDelta
Compiled
Time-`n` share of an all-time failure budget.
noncomputable def allTimeTelescopingDelta (delta : Real) (n : Nat) : Real
theorem
BanditRLProof.OFUL.allTimeTelescopingDelta_eq_div
Compiled
Display the scheduled budget as `delta / ((n+1)(n+2))`.
theorem allTimeTelescopingDelta_eq_div (delta : Real) (n : Nat) : allTimeTelescopingDelta delta n = delta / (((n + 1 : Nat) : Real) * ((n + 2 : Nat) : Real))
theorem
BanditRLProof.OFUL.allTimeTelescopingDelta_pos
Compiled
Positive outer budget gives positive timewise budgets.
theorem allTimeTelescopingDelta_pos {delta : Real} (hdelta : 0 < delta) (n : Nat) : 0 < allTimeTelescopingDelta delta n
theorem
BanditRLProof.OFUL.allTimeTelescopingDelta_le_one
Compiled
If `delta <= 1`, every timewise budget is also at most one.
theorem allTimeTelescopingDelta_le_one {delta : Real} (hdelta : 0 < delta) (hdelta_one : delta <= 1) (n : Nat) : allTimeTelescopingDelta delta n <= 1
theorem
BanditRLProof.OFUL.tsum_ofReal_allTimeTelescopingDelta
Compiled
The `ENNReal` scheduled failure budgets sum exactly to `ofReal delta`.
theorem tsum_ofReal_allTimeTelescopingDelta {delta : Real} (hdelta : 0 <= delta) : ∑' n, ENNReal.ofReal (allTimeTelescopingDelta delta n) = ENNReal.ofReal delta
def
BanditRLProof.OFUL.allTimeTelescopingScalarRidgeConfidenceFailureSet
Compiled
Failure set for the exact telescoping all-time confidence schedule.
def allTimeTelescopingScalarRidgeConfidenceFailureSet {Omega : Type u} {Feature : Type v} [Fintype Feature] [DecidableEq Feature] (lambda : Real) (thetaStar : Feature -> Real) (S : Real) (feature : Nat -> Omega -> Feature -> Real) (response : Nat -> Omega -> Real) (R delta : Real) : Set Omega
theorem
BanditRLProof.OFUL.measure_allTimeTelescopingScalarRidgeConfidenceFailureSet_le
Compiled
All-time scalar-ridge confidence under the exact telescoping failure schedule. The result is one countable event on one process, not a family of horizon-dependent generated algorithms.
theorem measure_allTimeTelescopingScalarRidgeConfidenceFailureSet_le {Omega : Type u} {Feature : Type v} [mOmega : MeasurableSpace Omega] [StandardBorelSpace Omega] [Fintype Feature] [DecidableEq Feature] [Nonempty Feature] (mu : Measure Omega) [IsProbabilityMeasure mu] (lambda : Real) (hlambda : 0 < lambda) (thetaStar : Feature -> Real) (S : Real) (htheta : euclideanLength thetaStar <= S) (F : Filtration Nat mOmega) (feature : Nat -> Omega -> Feature -> Real) (response noise : Nat -> Omega -> Real) (R : Real) (hR : 0 < R) (projectionBound : EuclideanSpace Real Feature -> Nat -> Real) (hfeature : forall i j, StronglyMeasurable[F i] (fun omega => feature i omega j)) (hnoise : StronglyAdapted F (fun t omega => match t with | 0 => 0 | i + 1 => noise i omega)) (hprojectionBound_nonneg : forall theta i, 0 <= projectionBound theta i) (hprojectionBound : forall theta i omega, |dotProduct (WithLp.ofLp theta) (feature i omega)| <= projectionBound theta i) (hsubGaussian : forall i, HasCondSubgaussianMGF (F i) (F.le i) (noise i) (constantSquaredVarianceProxy R i) mu) (hresponse : forall omega i, response i omega = dotProduct thetaStar (feature i omega) + noise i omega) (delta : Real) (hdelta : 0 < delta) (hdelta_one : delta <= 1) : mu (allTimeTelescopingScalarRidgeConfidenceFailureSet lambda thetaStar S feature response R delta) <= ENNReal.ofReal delta