Struct SparseVectorIndex 

pub struct SparseVectorIndex { /* private fields */ }

Scored sparse-vector inverted index over a DataType::SparseVector column.

Implementations

impl SparseVectorIndex

pub async fn new( base_uri: &str, config: SparseVectorIndexConfig, ) -> Result<Self>

Open or initialize a sparse index at base_uri for the given config.

pub async fn build_from_batches( &mut self, batches: &[RecordBatch], progress: impl Fn(usize), ) -> Result<()>

Rebuild the index from already-scanned record batches (the storage backend’s view of the flushed vertex table). This is the canonical backfill path: the LanceDB-managed table is opened by the backend, not by a raw lance::Dataset::open (whose physical path differs). Uses segmented accumulation to stay within the memory limit.

pub async fn query_topk( &self, query: &[(u32, f32)], k: usize, ) -> Result<Vec<(Vid, f32)>>

Score the corpus against query ([(term_id, weight)]) by dot product and return the top k (Vid, score) pairs, highest score first.

P1 brute-force document-at-a-time: filter postings to the query terms, accumulate per-vid dot products, then drain a bounded min-heap. No MVCC/tombstone filtering happens here — the orchestration layer applies it, mirroring the dense vector_search path.

pub async fn apply_incremental_updates( &mut self, added: &HashMap<Vid, Vec<(u32, f32)>>, removed: &HashSet<Vid>, ) -> Result<()>

Apply incremental updates: drop removed VIDs from every posting, then add the new vertices’ (term, weight) pairs, and rewrite. Mirrors the set-membership inverted index’s load-modify-write semantics.

pub fn is_initialized(&self) -> bool

Returns true if the index dataset exists.

pub fn property(&self) -> &str

Returns the property name this index is built on.

Trait Implementations

impl Debug for SparseVectorIndex

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

Formats the value using the given formatter.