Lean module · Probability layer
BanditRLProof.ConcentrationConfidenceSchedule
# Summable confidence schedules This module records reusable deterministic schedules for countable confidence budgets. It contains no stochastic-process or algorithm assumptions.
Module map
Imports
No project-local imports.
Imported by
BanditRLProof, BanditRLProof.ConcentrationFintypeGeometricAllTime, BanditRLProof.Exp3PredictableRegretAllTime, BanditRLProof.Exp3RealizedPredictableVarianceAllTime
Declarations
Open an item to read its exact compact statement and source link. Detailed teaching notes are linked when registered.
def
BanditRLProof.Concentration.geometricConfidenceShare
Compiled
The geometric confidence schedule `delta / 2 / 2^n`.
noncomputable def geometricConfidenceShare (delta : Real) (n : Nat) : Real
theorem
BanditRLProof.Concentration.geometricConfidenceShare_pos
Compiled
Every geometric confidence share is positive when its outer budget is.
theorem geometricConfidenceShare_pos {delta : Real} (hdelta : 0 < delta) (n : Nat) : 0 < geometricConfidenceShare delta n
theorem
BanditRLProof.Concentration.tsum_ofReal_geometricConfidenceShare
Compiled
The ENNReal masses of the geometric confidence shares sum exactly to the outer budget.
theorem tsum_ofReal_geometricConfidenceShare {delta : Real} (hdelta : 0 <= delta) : ∑' n, ENNReal.ofReal (geometricConfidenceShare delta n) = ENNReal.ofReal delta
def
BanditRLProof.Concentration.telescopingConfidenceWeight
Compiled
Positive telescoping weight `1 / ((n+1)(n+2))`. Unlike the geometric weight, its reciprocal grows only polynomially in time.
noncomputable def telescopingConfidenceWeight (n : Nat) : Real
theorem
BanditRLProof.Concentration.telescopingConfidenceWeight_eq_sub
Compiled
The telescoping weight is a difference of consecutive reciprocals.
theorem telescopingConfidenceWeight_eq_sub (n : Nat) : telescopingConfidenceWeight n = 1 / (((n + 1 : Nat) : Real)) - 1 / (((n + 2 : Nat) : Real))
theorem
BanditRLProof.Concentration.sum_range_telescopingConfidenceWeight
Compiled
Exact finite partial sum of the telescoping weights.
theorem sum_range_telescopingConfidenceWeight (n : Nat) : (Finset.range n).sum telescopingConfidenceWeight = 1 - 1 / (((n + 1 : Nat) : Real))
theorem
BanditRLProof.Concentration.telescopingConfidenceWeight_nonneg
Compiled
Every telescoping confidence weight is nonnegative.
theorem telescopingConfidenceWeight_nonneg (n : Nat) : 0 <= telescopingConfidenceWeight n
theorem
BanditRLProof.Concentration.hasSum_telescopingConfidenceWeight
Compiled
The telescoping weights sum exactly to one.
theorem hasSum_telescopingConfidenceWeight : HasSum telescopingConfidenceWeight 1
def
BanditRLProof.Concentration.telescopingConfidenceShare
Compiled
Time-`n` confidence share `delta / ((n+1)(n+2))`.
noncomputable def telescopingConfidenceShare (delta : Real) (n : Nat) : Real
theorem
BanditRLProof.Concentration.telescopingConfidenceShare_eq_div
Compiled
Display the telescoping share as the intended quotient.
theorem telescopingConfidenceShare_eq_div (delta : Real) (n : Nat) : telescopingConfidenceShare delta n = delta / (((n + 1 : Nat) : Real) * ((n + 2 : Nat) : Real))
theorem
BanditRLProof.Concentration.telescopingConfidenceShare_pos
Compiled
Every telescoping confidence share is positive when its outer budget is.
theorem telescopingConfidenceShare_pos {delta : Real} (hdelta : 0 < delta) (n : Nat) : 0 < telescopingConfidenceShare delta n
theorem
BanditRLProof.Concentration.tsum_ofReal_telescopingConfidenceShare
Compiled
The ENNReal masses of the telescoping confidence shares sum exactly to the outer budget.
theorem tsum_ofReal_telescopingConfidenceShare {delta : Real} (hdelta : 0 <= delta) : ∑' n, ENNReal.ofReal (telescopingConfidenceShare delta n) = ENNReal.ofReal delta