Struct PqCodebook

pub struct PqCodebook {
    pub m: usize,
    pub k: usize,
    pub dim: usize,
    pub centroids: Vec<f32>,
}

Product Quantization codebook.

Fields

m: usize

Number of subspaces.

k: usize

Number of centroids per subspace.

dim: usize

Original vector dimensionality.

centroids: Vec<f32>

Centroids: [subspace][centroid][sub_dim]. Flattened: length = m * k * (dim / m).

Implementations

impl PqCodebook

pub fn train(vectors: &[&[f32]], m: usize, k: usize, iterations: usize) -> Self

Train a codebook from a set of vectors using k-means.

• vectors: training data (each of length dim)
• m: number of subspaces
• k: centroids per subspace
• iterations: k-means iterations

pub fn sub_dim(&self) -> usize

Subdimension size.

pub fn encode(&self, vector: &[f32]) -> Vec<u8>

Encode a vector into PQ codes.

pub fn decode(&self, codes: &[u8]) -> Vec<f32>

Decode PQ codes back to an approximate vector.

pub fn build_distance_table(&self, query: &[f32]) -> Vec<Vec<f32>>

Build asymmetric distance table for a query vector.

Returns a table of size [m][k] where table[sub][code] is the squared distance from the query subvector to that centroid.

pub fn asymmetric_distance(table: &[Vec<f32>], codes: &[u8]) -> f32

Compute asymmetric distance from query to a PQ code using precomputed table.

Trait Implementations

impl Clone for PqCodebook

fn clone(&self) -> PqCodebook

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for PqCodebook

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.