Struct VectorQuantizer 

pub struct VectorQuantizer { /* private fields */ }

Vector quantizer for compressing and decompressing vectors.

Implementations

impl VectorQuantizer

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

Create a new vector quantizer.

The quantizer is created in an untrained state. Before calling quantize(), you must call train() with a representative set of vectors (unless the method is QuantizationMethod::None, which requires no training).

Arguments

• method - The quantization method to use.
• dimension - The expected vector dimension. All vectors passed to quantize() must have this dimension.

pub fn train(&mut self, vectors: &[Vector]) -> Result<()>

Train the quantizer on a set of vectors.

For Scalar8Bit, this computes per-dimension min/max values from the training set so that vectors can be linearly mapped to the [0, 255] range.

For None, no training is performed and this method returns immediately.

For ProductQuantization, this method currently returns an error because product quantization is not yet implemented.

Arguments

• vectors - A representative set of vectors to learn quantization parameters from. Must not be empty for Scalar8Bit.

Errors

Returns an error if the training set is empty (for Scalar8Bit) or if the quantization method is not yet implemented (ProductQuantization).

pub fn quantize(&self, vector: &Vector) -> Result<QuantizedVector>

Quantize a vector to compressed representation.

pub fn dequantize(&self, quantized: &QuantizedVector) -> Result<Vector>

Dequantize a compressed vector back to full precision.

pub fn compression_ratio(&self) -> f32

Get the compression ratio achieved by this quantization method.

pub fn is_trained(&self) -> bool

Check if the quantizer is trained.