Struct ExplorationConfig 

pub struct ExplorationConfig {
    pub epsilon: f64,
}

Epsilon-greedy start-rung exploration: a fraction epsilon of requests start at a uniformly-random rung so that propensities are logged and IPS/SNIPS off-policy estimates are valid (Horvitz-Thompson 1952; SNIPS: Swaminathan & Joachims 2015). The bandit still observes all gate verdicts — including from exploration draws — so learning is uninterrupted.

Every request under this policy records policy.propensity in the trace: the probability the logging policy had of choosing the start rung it actually chose. Old traces (before this field was added) serialize byte-identically (propensity is omitted when None).

Absent (default) → deterministic policy — no exploration, no propensity recorded, byte-identical to today.

Fields

epsilon: f64

Fraction of requests routed to a uniformly-random start rung instead of the greedy choice. Must be finite and in (0, 0.5]; validated by Config::parse. A small epsilon (0.05–0.1) is usually enough: it keeps learning alive and makes IPS/SNIPS estimates valid at low cost in exploration waste.

Trait Implementations

impl Clone for ExplorationConfig

fn clone(&self) -> ExplorationConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ExplorationConfig

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ExplorationConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.