Struct BM25Scorer 

pub struct BM25Scorer { /* private fields */ }

BM25 scorer for a document collection

Implementations

impl BM25Scorer

pub fn new(config: BM25Config) -> Self

Create a new BM25 scorer

pub fn build<I, D, T>(documents: I, config: BM25Config) -> Self

where I: IntoIterator<Item = D>, D: IntoIterator<Item = T>, T: AsRef<str>,

Build the scorer from a collection of documents

pub fn avg_doc_len(&self) -> f32

Average document length, derived from running totals.

Computed on read so it can never drift out of sync with the corpus (avgdl changes for every term on every insert and delete).

pub fn config(&self) -> BM25Config

The scoring configuration this scorer was built with.

pub fn idf(&self, term: &str) -> f32

Get IDF for a term.

Computed lazily from the current (df, N) so it is always consistent with the live corpus; unknown terms use df = 0 (maximum IDF). Terms whose IDF falls below min_idf contribute nothing.

pub fn score<I, T>( &self, query_terms: I, doc_terms: &[T], doc_len: usize, ) -> f32

where I: IntoIterator<Item = T>, T: AsRef<str> + Hash + Eq,

Score a document for a query

pub fn score_with_tf( &self, query_terms: &[String], doc_tf: &HashMap<String, usize>, doc_len: usize, ) -> f32

Score a document given precomputed term frequencies

pub fn score_with_tf_u32( &self, query_terms: &[String], doc_tf: &HashMap<String, u32>, doc_len: usize, ) -> f32

Score a document directly from a u32-valued term-frequency map.

Identical math to score_with_tf but lets callers whose postings already store u32 frequencies (the inverted index) score without cloning the whole term_freqs map into a usize-valued copy on every query.

pub fn add_document<I, T>(&mut self, tokens: I)

where I: IntoIterator<Item = T>, T: AsRef<str>,

Update stats when adding a document

pub fn remove_document<'a, I>(&mut self, unique_terms: I, doc_len: usize)

where I: IntoIterator<Item = &'a str>,

Update stats when removing a document.

Inverse of add_document: pass the document’s unique terms and its token length. Keeps num_docs, total_len, and doc_freqs consistent so IDF and avgdl never drift under deletion, and drops terms whose document frequency reaches zero (no vocabulary leak).

pub fn stats(&self) -> BM25Stats

Get statistics