cuda_ml_knn_algo_ivfsq {cuda.ml}R Documentation

Build a specification for the "ivfsq" KNN query algorithm.

Description

Build a specification of the inverted-file-scalar-quantization KNN query algorithm, with all required parameters specified explicitly.

Usage

cuda_ml_knn_algo_ivfsq(
  nlist,
  nprobe,
  qtype = c("QT_8bit", "QT_4bit", "QT_8bit_uniform", "QT_4bit_uniform", "QT_fp16",
    "QT_8bit_direct", "QT_6bit"),
  encode_residual = FALSE
)

Arguments

nlist

Number of cells to partition dataset into.

nprobe

At query time, the number of cells used for approximate nearest neighbor search.

qtype

Quantizer type. Must be one of "QT_8bit", "QT_4bit", "QT_8bit_uniform", "QT_4bit_uniform", "QT_fp16", "QT_8bit_direct", "QT_6bit".

encode_residual

Whether to encode residuals.

Value

An object encapsulating all required parameters of the "ivfsq" KNN query algorithm.


[Package cuda.ml version 0.3.2 Index]