Spectral Clustering

Spectral clustering is a graph-based clustering technique that uses the eigenvalues of similarity matrices to identify clusters with complex, non-convex shapes.

#include <cuvs/cluster/spectral.hpp>

namespace cuvs::cluster::spectral

Parameters

struct params

#include <spectral.hpp>

Parameters for spectral clustering.

Public Members

int n_clusters

Number of clusters to find

int n_components

Number of eigenvectors to use for the spectral embedding (typically equal to n_clusters)

int n_init

Number of k-means runs with different centroid seeds

int n_neighbors

Number of nearest neighbors for constructing the connectivity graph

float tolerance

Tolerance for the eigenvalue solver

raft::random::RngState rng_state = {0}

Random number generator state for reproducibility

Spectral Clustering

void fit_predict(

raft::resources const &handle,

params config,

raft::device_coo_matrix_view<float, int, int, int> connectivity_graph,

raft::device_vector_view<int, int> labels

)

Perform spectral clustering on a connectivity graph.

#include <cuvs/cluster/spectral.hpp>
#include <cuvs/preprocessing/spectral_embedding.hpp>

raft::resources handle;

// Create connectivity graph from data
auto graph = raft::make_device_coo_matrix<float>(handle, n_samples, n_samples);
cuvs::preprocessing::spectral_embedding::params embed_params;
embed_params.n_neighbors = 15;
cuvs::preprocessing::spectral_embedding::helpers::create_connectivity_graph(
    handle, embed_params, X.view(), graph);

// Configure and run spectral clustering
cuvs::cluster::spectral::params params;
params.n_clusters = 5;
params.n_components = 5;
params.n_neighbors = 15;
params.n_init = 10;

auto labels = raft::make_device_vector<int>(handle, n_samples);
cuvs::cluster::spectral::fit_predict(handle, params, graph.view(), labels.view());

Parameters:

• handle – [in] RAFT resource handle

• config – [in] Spectral clustering parameters

• connectivity_graph – [in] Sparse COO matrix representing connectivity between data points

• labels – [out] Device vector of size n_samples to store cluster assignments (0 to n_clusters-1)

void fit_predict(

raft::resources const &handle,

params config,

raft::device_coo_matrix_view<double, int, int, int> connectivity_graph,

raft::device_vector_view<int, int> labels

)

Perform spectral clustering on a connectivity graph.

#include <cuvs/cluster/spectral.hpp>
#include <cuvs/preprocessing/spectral_embedding.hpp>

raft::resources handle;

// Create connectivity graph from data
auto graph = raft::make_device_coo_matrix<double>(handle, n_samples, n_samples);
cuvs::preprocessing::spectral_embedding::params embed_params;
embed_params.n_neighbors = 15;
cuvs::preprocessing::spectral_embedding::helpers::create_connectivity_graph(
    handle, embed_params, X_double.view(), graph);

// Configure and run spectral clustering
cuvs::cluster::spectral::params params;
params.n_clusters = 5;
params.n_components = 5;
params.n_neighbors = 15;
params.n_init = 10;

auto labels = raft::make_device_vector<int>(handle, n_samples);
cuvs::cluster::spectral::fit_predict(handle, params, graph.view(), labels.view());

Parameters:

• handle – [in] RAFT resource handle

• config – [in] Spectral clustering parameters

• connectivity_graph – [in] Sparse COO matrix representing connectivity between data points

• labels – [out] Device vector of size n_samples to store cluster assignments (0 to n_clusters-1)