Struct PqEncoder 

pub struct PqEncoder { /* private fields */ }

A product quantizer with pre-trained codebooks.

The codebook codebooks[m][k] is the k-th centroid vector for sub-space m, with length config.subspace_dim().

Implementations

impl PqEncoder

pub fn new_random(config: PqConfig) -> Self

Create a PqEncoder with randomly initialised codebooks using a deterministic LCG so tests are reproducible.

pub fn encode(&self, vector: &[f32]) -> Result<Vec<usize>, String>

Encode a vector into M centroid indices (one per sub-space).

Returns an error if vector.len() != config.dimension.

pub fn decode(&self, codes: &[usize]) -> Result<Vec<f32>, String>

Decode M centroid indices back to an approximate reconstructed vector.

Returns an error if codes.len() != config.num_subspaces or any code index is out of bounds.

pub fn asymmetric_distance( &self, query: &[f32], codes: &[usize], ) -> Result<f32, String>

Compute the asymmetric distance between a query vector and encoded codes.

The asymmetric distance is the sum of squared Euclidean distances between each query sub-vector and its assigned centroid.

Returns an error if query.len() != config.dimension or codes are invalid.

pub fn config(&self) -> &PqConfig

Return a reference to the encoder’s configuration.