Struct IvfIndex 

pub struct IvfIndex {
    pub vectors: VecStore,
    pub original_ids: Vec<u32>,
    pub cluster_starts: Vec<u32>,
    pub tag_clusters: Vec<Vec<u16>>,
    pub dim: usize,
    pub n_clusters: usize,
    /* private fields */
}

IVF² index: geometric clusters × per-cluster tag posting lists.

Fields

vectors: VecStore

Reordered vectors (contiguous per cluster).

original_ids: Vec<u32>

Mapping: reordered_id → original_id.

cluster_starts: Vec<u32>

Cluster boundaries: cluster c spans [cluster_starts[c]..cluster_starts[c+1]).

tag_clusters: Vec<Vec<u16>>

Per-tag list of clusters containing matching vectors.

dim: usize

Vector dimensionality.

n_clusters: usize

Number of clusters.

Implementations

impl IvfIndex

pub fn build( base: &VecStore, base_meta: &SpMat, assignments: &[u16], n: usize, dim: usize, n_clusters: usize, ) -> Self

Build IVF² index from clustered vectors and metadata.

Reorders vectors by cluster, sorts within each cluster by most popular tag (tag-affinity sort), and builds per-cluster tag posting lists.

pub fn batch_search_mqcb( &self, queries: &QueryStore<'_>, nq: usize, query_tags: &[Vec<usize>], query_binary: &[Vec<u64>], query_top_clusters: &[Vec<usize>], binary: &BinaryStore, k: usize, ef: usize, n_probe: usize, binary_rerank: usize, ) -> Vec<Vec<u32>>

MQCB: processes queries grouped by cluster for L3 cache reuse.