Struct LegacyCompositePriorTransform 

pub struct LegacyCompositePriorTransform {
    pub alpha: f64,
    pub beta: f64,
    pub base_rate: Option<f64>,
}

Historical score transform that combines several dependent BM25-derived signals. It is intentionally named Legacy so callers do not mistake the output for a probability-model contract.

Fields

alpha: f64

beta: f64

base_rate: Option<f64>

Optional corpus-level base rate. None is equivalent to 0.5 (logit = 0, i.e. no base-rate correction).

Implementations

impl LegacyCompositePriorTransform

pub fn new(alpha: f64, beta: f64, base_rate: Option<f64>) -> ScoringResult<Self>

pub fn score_signal(&self, score: f64) -> f64

Bounded monotone score signal: sigma(alpha * (score - beta)).

This value is not a normalized P(score | relevant) likelihood.

pub fn tf_prior(tf: f64) -> f64

Term-frequency prior (Eq. 25): P_tf(tf) = 0.2 + 0.7 * min(1, tf / 10).

pub fn norm_prior(doc_len_ratio: f64) -> f64

Document-length normalisation prior (Eq. 26): P_norm(r) = 0.3 + 0.6 * (1 - min(1, |r - 0.5| * 2)), peaks at 0.9 when r = 0.5, floor 0.3 outside [0, 1].

pub fn composite_prior(tf: f64, doc_len_ratio: f64) -> f64

Composite prior (Eq. 27): clamp(0.7 * P_tf + 0.3 * P_norm, 0.1, 0.9).

pub fn no_match_floor() -> f64

Historical no-match floor obtained from zero term frequency and the document-length normalization floor. This is a ranking-policy value, not a corpus relevance prior.

pub fn combined_score( score_signal: f64, prior: f64, base_rate: Option<f64>, ) -> f64

Combine a bounded score signal and log-odds biases using the legacy two-stage probability-space arithmetic.

Without base_rate: P = L*p / (L*p + (1-L)*(1-p)).

With base_rate (the second update is equivalent to adding logit(base_rate) in log-odds space): Step 1: p1 = L*p / (L*p + (1-L)*(1-p)) Step 2: P = p1*br / (p1*br + (1-p1)*(1-br)).

pub fn transform_score(&self, score: f64, tf: f64, doc_len_ratio: f64) -> f64

Convert a BM25 score to the legacy bounded ranking score. tf is term frequency and doc_len_ratio is doc_length / avg_doc_length.

pub fn heuristic_upper_bound(&self, bm25_upper_bound: f64, p_max: f64) -> f64

Monotone upper bound for this transform, given a BM25 upper bound and an upper bound on the composite bias.

Trait Implementations

impl Clone for LegacyCompositePriorTransform

fn clone(&self) -> LegacyCompositePriorTransform

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for LegacyCompositePriorTransform

impl Debug for LegacyCompositePriorTransform

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

Formats the value using the given formatter.

impl Default for LegacyCompositePriorTransform

fn default() -> Self

Returns the “default value” for a type.