2.6. Online.Bandit.Algorithms.Regret.BayesRegretTS
Bayesian regret of Thompson sampling
This file provides a Bayesian regret upper bound (integral_regret_le) for Thompson sampling under
the assumption (among others) that it has the correct prior over environments.
The Bayesian regret upper bound relies on a clipped upper confidence bound whose definition and properties are also given in this file.
Main definitions
-
ucb A R l u σ2 δ a n: clipped upper confidence bound used in the regret analysis of Thompson sampling for a sequence of actionsA : ℕ → Ω → Fin K, rewardsR : ℕ → Ω → ℝ, reward lower boundl : ℝ, reward upper boundu : ℝ, sub-Gaussian variance proxyσ2 : ℝ, confidence parameterδ : ℝ, actiona : Fin K, and timen : ℕ. -
ucb' n h l u σ2 δ a: clipped upper confidence bound for actiona : Fin Kat timen : ℕgiven the historyh : Iic n → Fin K × ℝ(rather than the entire sequences of actions and rewards).
Main results
-
integral_regret_le: if Thompson sampling has the correct prior over environments and every environment hasKactions, each of which has a corresponding reward betweenlanduthat is sub-Gaussian with variance proxyσ2after its mean is subtracted, then the Bayesian regret at timenis at most(2 * K + 1) * (u - l) + 8 * √(σ2 * K * n * Real.log n).
Module LeanMachineLearning.Online.Bandit.Algorithms.Regret.BayesRegretTS contains 15 exposed declarations.
-
Bandits.ClippedUCB.ucb -
Bandits.ClippedUCB.ucb_zero -
Bandits.ClippedUCB.ucb_mem_Icc -
Bandits.ClippedUCB.measurable_ucb -
Bandits.ClippedUCB.measurable_uncurry_ucb_comp -
Bandits.ClippedUCB.integrable_uncurry_ucb_comp -
Bandits.ClippedUCB.ucb' -
Bandits.ClippedUCB.measurable_uncurry_ucb' -
Bandits.ClippedUCB.ucb_succ_eq_ucb' -
Bandits.ClippedUCB.sum_ucb_sub_mean_le -
Bandits.ClippedUCB.integral_sum_range_actionMean_bestAction_sub_ucb_bestAction_le -
Bandits.ClippedUCB.integral_sum_range_ucb_action_sub_actionMean_action_le -
Bandits.TS.integral_ucb_action_eq_integral_ucb_bestAction -
Bandits.TS.integral_regret_eq_add -
Bandits.TS.integral_regret_le