Struct OptimizedProductQuantizer 

pub struct OptimizedProductQuantizer { /* private fields */ }

Optimized Product Quantization (OPQ)

OPQ extends PQ by learning a rotation matrix that minimizes quantization error. It achieves better accuracy than standard PQ at the same compression ratio.

Implementations

impl OptimizedProductQuantizer

pub fn new(dimension: usize, num_subquantizers: usize, bits: u8) -> Result<Self>

Create a new OPQ quantizer

Arguments

• dimension - Vector dimension (must be divisible by num_subquantizers)
• num_subquantizers - Number of sub-quantizers
• bits - Bits per code (typically 8)

pub fn standard(dimension: usize) -> Result<Self>

Create standard OPQ with 8 sub-quantizers and 8 bits

pub fn train( &mut self, vectors: &[Vec<f32>], max_iterations: usize, rotation_iterations: usize, ) -> Result<()>

Train OPQ with rotation learning

Uses iterative optimization: alternate between learning rotation and PQ codebooks

Arguments

• vectors - Training vectors
• max_iterations - Max iterations for PQ k-means
• rotation_iterations - Iterations for rotation learning (typically 5-10)

pub fn quantize(&self, vector: &[f32]) -> Result<PQCode>

Quantize a vector

pub fn dequantize(&self, code: &PQCode) -> Result<Vec<f32>>

Dequantize a code back to vector

pub fn asymmetric_distance(&self, query: &[f32], code: &PQCode) -> Result<f32>

Compute asymmetric distance (query is not quantized)

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

Compute distance table for fast batch queries

pub fn distance_from_table(&self, table: &[Vec<f32>], code: &PQCode) -> f32

Fast distance using precomputed table

pub fn compression_ratio(&self) -> f32

Get compression ratio

pub fn is_trained(&self) -> bool

Check if trained

pub fn inner_pq(&self) -> &ProductQuantizer

Get the underlying PQ (for testing)

Trait Implementations

impl Clone for OptimizedProductQuantizer

fn clone(&self) -> OptimizedProductQuantizer

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for OptimizedProductQuantizer

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for OptimizedProductQuantizer

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for OptimizedProductQuantizer

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.