Lean module · Foundations
BanditRLProof.HeavyTailTuning
Algebraic tuning for the sample-index threshold; all exponents remain real.
Module map
Imports
BanditRLProof.HeavyTailPowerSum, BanditRLProof.Algorithms.HeavyTailUCB
Imported by
BanditRLProof, BanditRLProof.HeavyTailClippedConfidence, BanditRLProof.HeavyTailClippedScheduled, BanditRLProof.HeavyTailGapThreshold, BanditRLProof.HeavyTailScheduledConfidence, BanditRLProof.HeavyTailSourceConfidence, BanditRLProof.HeavyTailTailSum
Declarations
Open an item to read its exact compact statement and source link. Detailed teaching notes are linked when registered.
theorem
BanditRLProof.HeavyTail.power_threshold_bias_term
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.power_threshold_bias_termReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem power_threshold_bias_term (u c a ε x : ℝ) (hc : 0 < c) (hx : 0 < x) (haε : a * ε = 1-a) : u / (c * x^a)^ε = (u / c^ε) * x^(a-1)
theorem
BanditRLProof.HeavyTail.power_threshold_bias_sum
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.power_threshold_bias_sumReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem power_threshold_bias_sum (u c a ε : ℝ) (hu : 0 ≤ u) (hc : 0 < c) (ha : 0 < a) (ha1 : a ≤ 1) (haε : a * ε = 1-a) (n : ℕ) : (∑ s ∈ Finset.range n, u / (c * ((s : ℝ)+1)^a)^ε) ≤ (u / c^ε) * ((n : ℝ)^a / a)
theorem
BanditRLProof.HeavyTail.sampleThreshold_factor
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.sampleThreshold_factorReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem sampleThreshold_factor (ε u : ℝ) (hu : 0 ≤ u) (t s : ℕ) : sampleThreshold ε u t s = (u / confidenceLog t)^(1/(1+ε)) * ((s : ℝ)+1)^(1/(1+ε))
theorem
BanditRLProof.HeavyTail.sampleThreshold_bias_sum
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.sampleThreshold_bias_sumReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem sampleThreshold_bias_sum (ε u : ℝ) (hε : 0 ≤ ε) (hu : 0 < u) (t n : ℕ) : (∑ s ∈ Finset.range n, u / (sampleThreshold ε u t s)^ε) ≤ (u / ((u / confidenceLog t)^(1/(1+ε)))^ε) * ((n : ℝ)^(1/(1+ε)) / (1/(1+ε)))
theorem
BanditRLProof.HeavyTail.power_scale_bias
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.power_scale_biasReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem power_scale_bias (u L a ε : ℝ) (hu : 0 < u) (hL : 0 < L) (haε : a * ε = 1-a) : u / ((u/L)^a)^ε = u^a * L^(1-a)
theorem
BanditRLProof.HeavyTail.power_bias_normalization
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.power_bias_normalizationReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem power_bias_normalization (u L a N : ℝ) (hL : 0 < L) (hN : 0 < N) : (u^a * L^(1-a)) * (N^a / a) / N = (1/a) * u^a * (L/N)^(1-a)
theorem
BanditRLProof.HeavyTail.sampleThreshold_bias_average
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.sampleThreshold_bias_averageReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem sampleThreshold_bias_average (ε u : ℝ) (hε : 0 ≤ ε) (hu : 0 < u) (t n : ℕ) (hn : 0 < n) : (∑ s ∈ Finset.range n, u / (sampleThreshold ε u t s)^ε) / n ≤ (1+ε) * u^(1/(1+ε)) * (confidenceLog t / n)^(ε/(1+ε))
theorem
BanditRLProof.HeavyTail.threshold_scale_identity
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.threshold_scale_identityReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem threshold_scale_identity (u L N a : ℝ) (hu : 0 ≤ u) (hL : 0 < L) (hN : 0 < N) : (u*N/L)^a * L = N * (u^a * (L/N)^(1-a))
theorem
BanditRLProof.HeavyTail.threshold_variance_identity
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.threshold_variance_identityReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem threshold_variance_identity (u L N ε : ℝ) (hu : 0 < u) (hL : 0 < L) (hN : 0 < N) (hp : 0 < 1+ε) : N*u*((u*N/L)^(1/(1+ε)))^(1-ε)*L = ((u*N/L)^(1/(1+ε))*L)^2
theorem
BanditRLProof.HeavyTail.confidenceLog_pos
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.confidenceLog_posReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem confidenceLog_pos (t : ℕ) : 0 < confidenceLog t
theorem
BanditRLProof.HeavyTail.sampleThreshold_pos
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.sampleThreshold_posReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem sampleThreshold_pos (ε u : ℝ) (hu : 0 < u) (t s : ℕ) : 0 < sampleThreshold ε u t s
theorem
BanditRLProof.HeavyTail.sampleThreshold_le_terminal
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.sampleThreshold_le_terminalReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem sampleThreshold_le_terminal (ε u : ℝ) (hε : 0 ≤ ε) (hu : 0 ≤ u) (t n s : ℕ) (hs : s < n) : sampleThreshold ε u t s ≤ (u*n/confidenceLog t)^(1/(1+ε))
theorem
BanditRLProof.HeavyTail.sampleThreshold_variance_sum
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.sampleThreshold_variance_sumReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem sampleThreshold_variance_sum (ε u : ℝ) (hε0 : 0 ≤ ε) (hε : ε ≤ 1) (hu : 0 < u) (t n : ℕ) : (∑ s ∈ Finset.range n, u * (sampleThreshold ε u t s)^(1-ε)) ≤ n*u*((u*n/confidenceLog t)^(1/(1+ε)))^(1-ε)
theorem
BanditRLProof.HeavyTail.tuned_radius_le
Compiled
No declaration docstring is present; use the chapter context and exact statement below.
Used in these reading views: Bandit Book
1. Finite bandits, traces, and regret
Canonical node identity
declaration:BanditRLProof.HeavyTail.tuned_radius_leReading membership is not a proof dependency. Exact assumptions remain in the Lean statement.
theorem tuned_radius_le (ε u : ℝ) (hε0 : 0 ≤ ε) (hε : ε ≤ 1) (hu : 0 < u) (t n : ℕ) (hn : 0 < n) : ((∑ s ∈ Finset.range n, u / (sampleThreshold ε u t s)^ε) + (2 * Real.sqrt ((∑ s ∈ Finset.range n, u * (sampleThreshold ε u t s)^(1-ε)) * confidenceLog t) + 2 * (u*n/confidenceLog t)^(1/(1+ε)) * confidenceLog t)) / n ≤ confidenceRadius ε u t n