Struct Quantizer 

pub struct Quantizer { /* private fields */ }

Product Quantizer that compresses high-dimensional vectors.

Implementations

impl Quantizer

pub fn new(config: QuantizerConfig) -> Self

Create a new, un-trained quantizer with the given configuration.

pub fn train(&mut self, data: &[Vec<f32>]) -> Result<(), QuantizerError>

Train the quantizer using k-means on the provided data vectors.

All vectors must have the same dimensionality, which must be divisible by n_subspaces.

pub fn encode(&self, vector: &[f32]) -> Result<QuantizedVector, QuantizerError>

Encode a single vector into a QuantizedVector.

pub fn decode( &self, qv: &QuantizedVector, ) -> Result<ReconstructedVector, QuantizerError>

Decode a QuantizedVector back to an approximate vector.

pub fn encode_batch( &self, vectors: &[Vec<f32>], ) -> Result<Vec<QuantizedVector>, QuantizerError>

Encode a batch of vectors.

pub fn is_trained(&self) -> bool

Return true if the quantizer has been trained.

pub fn compression_ratio(&self, original_dim: usize) -> f32

Compute the compression ratio: bytes of original / bytes of compressed.

Original: original_dim * 4 bytes (f32). Compressed: n_subspaces bytes (one u8 code per sub-space).

pub fn codebook_count(&self) -> usize

Return the number of codebooks (equals n_subspaces after training).

Trait Implementations

impl Debug for Quantizer

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

Formats the value using the given formatter.