Implementation:Rapidsai Cuml DistanceType Enum

Knowledge Sources	Rapidsai_Cuml
Domains	Machine_Learning, Distance_Metrics
Last Updated	2026-02-08 12:00 GMT

Overview

Defines an enumeration of distance metric types used throughout the cuML library for specifying how distances between data points are computed.

Description

The ML::distance::DistanceType enum class provides a comprehensive set of distance and similarity metrics commonly used in machine learning algorithms. It includes Euclidean (L2), Manhattan (L1), cosine similarity, Minkowski (Lp), and many others. Each enumerator is assigned a unique integer value, enabling efficient selection of distance computation strategies at runtime. The enum resides in the ML::distance namespace and is referenced by clustering, nearest-neighbor, silhouette score, and pairwise distance APIs across cuML.

The special value Precomputed = 100 indicates that a precomputed distance matrix is provided rather than raw feature data, which avoids redundant computation when distances have already been calculated.

Usage

Use this enum whenever a cuML C++ API requires a distance metric parameter. It is passed to functions such as silhouette_score, pairwise_distance, and various clustering routines to select the desired distance computation. Import the header and specify the desired metric when calling these functions.

Code Reference

Source Location

• Repository: Rapidsai_Cuml
• File: cpp/include/cuml/common/distance_type.hpp

Signature

namespace ML::distance {

enum class DistanceType {
  L2Expanded          = 0,
  L2SqrtExpanded      = 1,
  CosineExpanded      = 2,
  L1                  = 3,
  L2Unexpanded        = 4,
  L2SqrtUnexpanded    = 5,
  InnerProduct        = 6,
  Linf                = 7,
  Canberra            = 8,
  LpUnexpanded        = 9,
  CorrelationExpanded = 10,
  JaccardExpanded     = 11,
  HellingerExpanded   = 12,
  Haversine           = 13,
  BrayCurtis          = 14,
  JensenShannon       = 15,
  HammingUnexpanded   = 16,
  KLDivergence        = 17,
  RusselRaoExpanded   = 18,
  DiceExpanded        = 19,
  BitwiseHamming      = 20,
  Precomputed         = 100
};

}  // end namespace ML::distance

Import

#include <cuml/common/distance_type.hpp>

I/O Contract

Inputs

Name	Type	Required	Description
(N/A -- this is an enum definition)

Outputs

Name	Type	Description
DistanceType	enum class (int)	An enumerator value representing the chosen distance metric.

Enum Values

Enumerator	Value	Description
L2Expanded	0	Squared L2 (Euclidean) distance, expanded form
L2SqrtExpanded	1	L2 (Euclidean) distance (square root of L2Expanded)
CosineExpanded	2	Cosine distance (1 - cosine similarity)
L1	3	Manhattan (L1) distance
L2Unexpanded	4	Squared L2 distance, unexpanded form
L2SqrtUnexpanded	5	L2 distance, unexpanded form
InnerProduct	6	Inner (dot) product
Linf	7	Chebyshev (L-infinity) distance
Canberra	8	Canberra distance
LpUnexpanded	9	Minkowski (Lp) distance, unexpanded form
CorrelationExpanded	10	Correlation distance
JaccardExpanded	11	Jaccard distance
HellingerExpanded	12	Hellinger distance
Haversine	13	Haversine distance for geographical coordinates
BrayCurtis	14	Bray-Curtis distance
JensenShannon	15	Jensen-Shannon divergence
HammingUnexpanded	16	Hamming distance
KLDivergence	17	Kullback-Leibler divergence
RusselRaoExpanded	18	Russell-Rao distance
DiceExpanded	19	Dice distance
BitwiseHamming	20	Bitwise Hamming distance
Precomputed	100	Indicates a precomputed distance matrix is provided

Usage Examples

#include <cuml/common/distance_type.hpp>
#include <cuml/metrics/metrics.hpp>
#include <raft/core/handle.hpp>

void compute_pairwise() {
    raft::handle_t handle;

    // Assume x and y are device pointers to feature matrices,
    // dist is a device pointer to the output distance matrix.
    float* x = /* ... */;
    float* y = /* ... */;
    float* dist = /* ... */;
    int m = 100, n = 200, k = 50;

    // Compute pairwise L2 (Euclidean) distances
    ML::Metrics::pairwise_distance(
        handle, x, y, dist, m, n, k,
        ML::distance::DistanceType::L2SqrtExpanded,
        true  // row-major
    );
}

Related Pages

• Environment:Rapidsai_Cuml_CUDA_GPU