Struct SieveCollection 

pub struct SieveCollection { /* private fields */ }

A collection of specialized HNSW subindices, one per stable predicate.

Each subindex is built over the subset of vectors that match its predicate. Smaller datasets mean a lower sub_m is sufficient, keeping memory overhead proportional to the per-predicate population rather than the global index.

Implementations

impl SieveCollection

pub fn new(sub_m: usize) -> Self

Create an empty collection. sub_m is the HNSW M parameter used for every subindex built by this collection.

pub fn build_subindex( &mut self, signature: PredicateSignature, vectors: &[(u32, Vec<f32>)], dim: usize, metric: DistanceMetric, ) -> Result<(), VectorError>

Build (or rebuild) a specialized subindex for signature from the provided (id, vector) pairs.

All vectors must have length dim. The subindex uses metric for distance computation.

Errors

Returns VectorError if any insertion fails (e.g. dimension mismatch).

pub fn has(&self, signature: &PredicateSignature) -> bool

Returns true if a subindex exists for the given signature.

pub fn get(&self, signature: &PredicateSignature) -> Option<&HnswIndex>

Returns a shared reference to the subindex for signature, or None.

pub fn drop(&mut self, signature: &PredicateSignature)

Remove the subindex for signature, freeing its memory.

pub fn signatures(&self) -> Vec<&PredicateSignature>

All predicate signatures currently held in this collection.