Struct BayesianBM25Params 

pub struct BayesianBM25Params {
    pub bm25: BM25Params,
    pub alpha: f64,
    pub beta: f64,
    pub base_rate: f64,
    pub calibration_tokens: f64,
    pub beta_slope: f64,
    pub sigma_slope: f64,
}

Fields

bm25: BM25Params

alpha: f64

beta: f64

base_rate: f64

Corpus relevance prior in [0, 1); zero means “not estimated”. The prior never enters the posterior transform (which matches Lucene’s BayesianScoreQuery exactly); it is metadata for fusion, where it converts posteriors to evidence and enters the fused score exactly once.

calibration_tokens: f64

Query length (analyzed term count) the calibration was fitted at. Zero disables query-length scaling, so hand-written and learner-fitted parameters apply verbatim.

beta_slope: f64

Fitted per-token slope of the sigmoid midpoint: raw BM25 sums grow with the number of query terms, and beta must track that scale for the posterior to stay in its linear region.

sigma_slope: f64

Fitted per-token slope of the score spread (1 / alpha).

Implementations

impl BayesianBM25Params

pub fn evidence_params(&self) -> Self

Parameters whose posterior equals this calibration’s prior-free evidence sigmoid(alpha * (raw - beta) - logit(base_rate)), expressed through the equivalent midpoint shift beta + logit(base_rate) / alpha. With no estimated prior the calibration is returned unchanged.

pub fn scaled_for_query_terms(&self, term_count: usize) -> Self

The calibration translated to a query with term_count analyzed terms. Raw BM25 query scores are sums over query terms, so both the midpoint and the spread of the matching-score distribution move with the term count; the estimator fits those slopes and this method applies them:

beta_q = beta + beta_slope * (q - q_ref) sigma_q = max(sigma_ref + sigma_slope * (q - q_ref), floor)

Parameters without a fitted reference length (or a zero term count) are returned unchanged. Scaling happens per query, never per document, so within-query ranking stays monotone in the raw score.

Trait Implementations

impl Clone for BayesianBM25Params

fn clone(&self) -> BayesianBM25Params

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for BayesianBM25Params

impl Debug for BayesianBM25Params

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

Formats the value using the given formatter.

impl Default for BayesianBM25Params

fn default() -> Self

Returns the “default value” for a type.

impl PartialEq for BayesianBM25Params

fn eq(&self, other: &BayesianBM25Params) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for BayesianBM25Params