Struct SparseInvertedIndex 

pub struct SparseInvertedIndex { /* private fields */ }

Inverted index over sparse vectors for a single (collection, field) pair.

Documents are addressed externally by their string ID and internally by a dense u32 so posting lists stay compact.

Implementations

impl SparseInvertedIndex

pub fn new() -> Self

Create an empty index.

pub fn insert(&mut self, doc_id: &str, vector: &SparseVector)

Insert or replace the sparse vector for doc_id.

Upsert semantics: any postings previously recorded for doc_id are removed before the new ones are appended, so re-indexing a document never duplicates its postings.

pub fn delete(&mut self, doc_id: &str) -> bool

Remove a document entirely. Returns true when it was present.

pub fn search(&self, query: &SparseVector, top_k: usize) -> Vec<SparseHit>

Top-k documents by dot product against query, descending.

Only the query’s non-zero dimensions are visited. Documents sharing no dimension with the query never enter the accumulator and are therefore excluded, as are documents whose weights cancel to exactly zero. Ties on score are broken by ascending doc_id so ordering is stable across runs and across a checkpoint round-trip.

pub fn doc_count(&self) -> usize

Number of documents currently indexed.

pub fn is_empty(&self) -> bool

Whether the index holds no documents.

pub fn documents(&self) -> Vec<(String, Vec<(u32, f32)>)>

Every indexed document as (doc_id, entries), sorted by doc_id so the checkpoint blob is byte-stable for a given logical state.

pub fn from_documents(documents: Vec<(String, Vec<(u32, f32)>)>) -> Self

Rebuild an index from checkpointed (doc_id, entries) pairs.

Entries are already normalized (sorted, deduplicated, finite) by SparseVector at insert time; malformed rows are dropped rather than failing the whole restore, since a partially readable index is strictly better than none.

Trait Implementations

impl Default for SparseInvertedIndex

fn default() -> Self

Returns the “default value” for a type.