Struct LightPQTrainingParameters 

pub struct LightPQTrainingParameters { /* private fields */ }

Implementations

impl LightPQTrainingParameters

pub fn new(ncenters: usize, lloyds_reps: usize) -> Self

Construct a new light-weight PQ trainer.

Trait Implementations

impl TrainQuantizer for LightPQTrainingParameters

fn train<R, C>( &self, data: MatrixView<'_, f32>, schema: ChunkOffsetsView<'_>, parallelism: Parallelism, rng_builder: &R, cancelation: &C, ) -> Result<Self::Quantizer, Self::Error>

where R: RngBuilder<usize> + Sync, C: Cancelation + Sync,

Perform product quantization training on the provided training set and return a SimplePivots containing the result of kmeans clustering on each partition.

Panics

Panics if data.nrows() != schema.dim().

Errors

An error type is returned under the following circumstances:

• A cancellation request is received. This case can be queried by calling was_canceled on the returned PQTrainingError.
• NaN or infinities are observed during the training process.

type Quantizer = SimplePivots

type Error = PQTrainingError