Struct QueryPoolVectorScoreOperator 

pub struct QueryPoolVectorScoreOperator {
    pub query_vector: Vec<f32>,
    pub k: usize,
    pub field: String,
    pub base_rate: f64,
    pub split: RelevantSampleSplit,
}

Query-pool vector score transform.

Fits the likelihood-ratio calibration from the retrieved pool at query time: the head of the sorted distance distribution (per RelevantSampleSplit) estimates the relevant density f_R, the tail estimates the background density f_G, and each candidate’s posterior is sigmoid(log(f_R(d) / f_G(d)) + logit(base_rate)) via VectorProbabilityTransform. An uninformative pool (too small, zero spread, or no head/tail separation) yields the prior for every candidate instead of fabricating discrimination. Because the same selected pool supplies both pseudo-classes, this is an unsupervised ranking transform, not a reusable calibrated-probability model. Use uqa_scoring::VectorCalibrationModel for the latter contract.

Fields

query_vector: Vec<f32>

k: usize

field: String

base_rate: f64

Relevance prior folded into the posterior. The default 0.5 contributes zero log-odds, so the output doubles as prior-free evidence for fusion-level priors.

split: RelevantSampleSplit

Implementations

impl QueryPoolVectorScoreOperator

pub fn new(query_vector: Vec<f32>, k: usize, field: impl Into<String>) -> Self

pub fn with_split(self, split: RelevantSampleSplit) -> Self

pub fn with_base_rate(self, base_rate: f64) -> Self

Trait Implementations

impl Operator for QueryPoolVectorScoreOperator

fn execute(&self, ctx: &ExecutionContext) -> OperatorResult

fn cost_estimate(&self, stats: &IndexStats) -> f64