Struct InvertedIndex 

pub struct InvertedIndex<E: StorageEngine> { /* private fields */ }

Inverted index for sparse vector similarity search.

Implementations

impl<E: StorageEngine> InvertedIndex<E>

pub const fn new(engine: E) -> Self

Create a new inverted index with the given storage engine.

pub fn engine(&self) -> &E

Get a reference to the storage engine.

pub fn insert( &self, collection: &str, vector_name: &str, point_id: PointId, vector: &[(u32, f32)], ) -> Result<(), VectorError>

Insert a sparse vector into the index.

Arguments

• collection - Collection name
• vector_name - Vector name within the collection
• point_id - Point ID
• vector - Sparse vector as (token_id, weight) pairs (must be sorted)

pub fn delete( &self, collection: &str, vector_name: &str, point_id: PointId, ) -> Result<bool, VectorError>

Delete a sparse vector from the index.

Returns

Returns Ok(true) if the vector was deleted, Ok(false) if it wasn’t indexed.

pub fn update( &self, collection: &str, vector_name: &str, point_id: PointId, vector: &[(u32, f32)], ) -> Result<(), VectorError>

Update a sparse vector in the index (delete + insert).

pub fn delete_collection(&self, collection: &str) -> Result<(), VectorError>

Delete all index data for a collection.

pub fn delete_vector( &self, collection: &str, vector_name: &str, ) -> Result<(), VectorError>

Delete all index data for a specific vector in a collection.

pub fn get_meta( &self, collection: &str, vector_name: &str, ) -> Result<InvertedIndexMeta, VectorError>

Get index metadata.

pub fn get_posting_list( &self, collection: &str, vector_name: &str, token_id: u32, ) -> Result<Option<PostingList>, VectorError>

Get a posting list for a specific token.

pub fn search_daat( &self, collection: &str, vector_name: &str, query: &[(u32, f32)], top_k: usize, scoring: ScoringFunction, ) -> Result<Vec<SearchResult>, VectorError>

Search using DAAT (Document-at-a-time) algorithm.

This is exact scoring that traverses all posting lists to compute the full similarity score for each candidate document.

Arguments

• collection - Collection name
• vector_name - Vector name
• query - Query vector as (token_id, weight) pairs
• top_k - Number of results to return
• scoring - Scoring function to use

pub fn search_wand( &self, collection: &str, vector_name: &str, query: &[(u32, f32)], top_k: usize, ) -> Result<Vec<SearchResult>, VectorError>

Search using WAND (Weak AND) algorithm.

This is an optimized top-k search that uses upper bound scores to skip documents that cannot make it into the result set.

Arguments

• collection - Collection name
• vector_name - Vector name
• query - Query vector as (token_id, weight) pairs
• top_k - Number of results to return

pub fn search_maxscore( &self, collection: &str, vector_name: &str, query: &[(u32, f32)], top_k: usize, ) -> Result<Vec<SearchResult>, VectorError>

Search using MaxScore algorithm (optimized WAND variant).

Similar to WAND but more aggressive at skipping low-scoring documents.