Struct PQConfig 

pub struct PQConfig {
    pub dimension: usize,
    pub num_segments: usize,
    pub num_centroids: usize,
    pub distance_metric: DistanceMetric,
    pub training_iterations: usize,
    pub seed: Option<u64>,
}

Configuration for Product Quantization.

Parameters

• dimension: The dimension of vectors to quantize
• num_segments: Number of subspaces (M). Must divide dimension evenly.
• num_centroids: Number of centroids per subspace (K). Typically 256 (8 bits per code).
• distance_metric: Distance metric for codebook training and distance computation

Memory Usage

• Codebooks: M × K × (D/M) × 4 bytes
• Per-vector codes: M × ceil(log2(K)/8) bytes

For typical settings (D=128, M=8, K=256):

• Codebooks: 8 × 256 × 16 × 4 = 128KB
• Per-vector: 8 bytes (compression ratio: 64x)

Fields

dimension: usize

Dimension of input vectors.

num_segments: usize

Number of subspaces (segments).

num_centroids: usize

Number of centroids per subspace.

distance_metric: DistanceMetric

Distance metric for training and search.

training_iterations: usize

Number of training iterations for k-means.

seed: Option<u64>

Random seed for reproducible training.

Implementations

impl PQConfig

pub fn new(dimension: usize, num_segments: usize) -> Self

Create a new PQ configuration.

Arguments

• dimension: The dimension of vectors to quantize
• num_segments: Number of subspaces. Must divide dimension evenly.

Defaults

• num_centroids: 256 (8-bit codes)
• distance_metric: Euclidean
• training_iterations: 25
• seed: None (non-deterministic)

Panics

Panics if num_segments is 0 or doesn’t divide dimension evenly.

pub const fn with_num_centroids(self, k: usize) -> Self

Set the number of centroids per subspace.

Common values:

• 256 (8-bit codes, default)
• 65536 (16-bit codes, more accurate but larger codebooks)

pub const fn with_distance_metric(self, metric: DistanceMetric) -> Self

Set the distance metric.

pub const fn with_training_iterations(self, iterations: usize) -> Self

Set the number of training iterations.

pub const fn with_seed(self, seed: u64) -> Self

Set the random seed for reproducible training.

pub const fn subspace_dimension(&self) -> usize

Get the dimension of each subspace.

pub fn bits_per_code(&self) -> usize

Calculate the number of bits per code.

Returns the number of bits needed to represent a centroid index.

pub fn bytes_per_code(&self) -> usize

Calculate bytes per encoded vector.

pub fn validate(&self) -> Result<(), VectorError>

Validate the configuration.

Errors

Returns an error if:

• dimension is 0
• num_segments is 0 or doesn’t divide dimension
• num_centroids is 0

pub fn compression_ratio(&self) -> f32

Calculate compression ratio compared to full f32 vectors.

Trait Implementations

impl Clone for PQConfig

fn clone(&self) -> PQConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for PQConfig

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

Formats the value using the given formatter.