Struct DiscountedThompsonSampling 

pub struct DiscountedThompsonSampling { /* private fields */ }

Thompson Sampling with exponential discounting (Qi+2023).

Extends standard Beta-Bernoulli Thompson Sampling with a forgetting factor that decays posterior parameters toward the prior on every step. This allows the algorithm to adapt in non-stationary environments where the best arm can change over time.

Used in irithyll’s AutoML module to select hyperparameter configurations when the optimal configuration may shift due to concept drift.

Implementations

impl DiscountedThompsonSampling

pub fn new(n_arms: usize, discount: f64) -> Self

Create a new DiscountedThompsonSampling bandit with the default uniform Beta(1, 1) prior and seed 42.

Panics

Panics if discount is not in (0, 1] or if n_arms is zero.

pub fn with_seed(n_arms: usize, discount: f64, seed: u64) -> Self

Create a new DiscountedThompsonSampling bandit with a custom RNG seed and the default Beta(1, 1) prior.

Panics

Panics if discount is not in (0, 1] or if n_arms is zero.

pub fn with_prior( n_arms: usize, discount: f64, alpha_prior: f64, beta_prior: f64, ) -> Self

Create a new DiscountedThompsonSampling bandit with a custom Beta(alpha, beta) prior for all arms and seed 42.

Panics

Panics if alpha_prior or beta_prior are not positive, if discount is not in (0, 1], or if n_arms is zero.

pub fn discount(&self) -> f64

The discount (forgetting) factor.

pub fn arm_alphas(&self) -> &[f64]

Per-arm posterior alpha parameters.

pub fn arm_betas(&self) -> &[f64]

Per-arm posterior beta parameters.

Trait Implementations

impl Bandit for DiscountedThompsonSampling

fn select_arm(&mut self) -> usize

Select the next arm to pull.

fn update(&mut self, arm: usize, reward: f64)

Update the bandit with observed reward for the given arm.

fn n_arms(&self) -> usize

Number of arms.

fn n_pulls(&self) -> u64

Total number of pulls across all arms.

fn reset(&mut self)

Reset to initial state (all counts and estimates zeroed).

fn arm_values(&self) -> &[f64]

Per-arm estimated reward values.

fn arm_counts(&self) -> &[u64]

Per-arm pull counts.

impl Clone for DiscountedThompsonSampling

fn clone(&self) -> DiscountedThompsonSampling

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for DiscountedThompsonSampling

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.