Struct ScalarQuantizationParameters 

pub struct ScalarQuantizationParameters { /* private fields */ }

Parameters controlling the generation of the scalar quantization Quantizer.

When performing scalar quantization, the mean of each dimension will be calculated and the dataset will be shifted around this mean.

Next, the standard deviation of each dimension will be computed and the maximum m found.

The dynamic range of the final compressed encoding will then span 2 * standard_deviations * m for each dimension symmetrically about the mean for each dimension. Values outside the spanned dynamic range will be clamped.

Implementations

impl ScalarQuantizationParameters

pub fn new(standard_deviations: Positive<f64>) -> Self

Construct a new quantizer with the given parameters.

Arguments

• standard_deviations: The number of maximal standard deviations to use for the encoding’s dynamic range. This number must be positive, and generally should be greater than 1.0.

  A good starting value is generally 2.0.

pub fn standard_deviations(&self) -> Positive<f64>

Return the current number of standard deviations being used to set the dynamic range.

pub fn train<T>(&self, data: MatrixView<'_, T>) -> ScalarQuantizer

where T: Copy + Into<f64> + Into<f32>,

Train a new ScalarQuantizer on the input training data.

The training algoritm works as follows:

1. The medoid of the training data is computed.

2. The standard deviation for each dimension is then calculated across all rows of the training set.

3. The maximum standard deviation s is computed and the dynamic range dyn of the quantizer is computed as dyn = 2.0 * self.standard_deviations() * s.

4. The quantizer is then constructed with scale = dyn / (2.pow(NBITS) - 1).

Complexity

This method is linear in the number of rows and columns in data.

Allocates

This method allocated memory on the order of data.ncols() (the dimensionality of the data).

Parallelism

This function is single threaded.

Trait Implementations

impl Default for ScalarQuantizationParameters

fn default() -> Self

Returns the “default value” for a type.