Expand description
In-memory inverted index for sparse vectors (learned sparse retrieval).
Maps each dimension to a posting list of (internal_doc_id, weight) pairs.
Scoring accumulates the dot product over only the query’s non-zero
dimensions, so cost is proportional to the query’s nnz times the average
posting-list length — never to the number of indexed documents.
Maintained synchronously on document write / delete and checkpoint- serializable so a reopen never needs a rebuild.
Structs§
- Sparse
Hit - A scored hit from the sparse inverted index.
- Sparse
Inverted Index - Inverted index over sparse vectors for a single
(collection, field)pair.