Struct Kmeans 

pub struct Kmeans<D: DistanceMetric = SquaredEuclidean> { /* private fields */ }

K-means clustering algorithm, generic over a distance metric.

The default metric is clump::SquaredEuclidean, preserving backward compatibility: Kmeans::new(k) works exactly as before.

To use a different metric, construct via Kmeans::with_metric:

use sheaf::cluster::{Kmeans, CosineDistance};

let km = Kmeans::with_metric(8, CosineDistance).with_seed(42);

Implementations

impl Kmeans<SquaredEuclidean>

pub fn new(k: usize) -> Self

Create a new K-means clusterer with the default squared Euclidean distance.

impl<D: DistanceMetric> Kmeans<D>

pub fn with_metric(k: usize, metric: D) -> Self

Create a new K-means clusterer with a custom distance metric.

pub fn with_max_iter(self, max_iter: usize) -> Self

Set maximum iterations.

pub fn with_tol(self, tol: f64) -> Self

Set convergence tolerance.

pub fn with_seed(self, seed: u64) -> Self

Set random seed for reproducibility.

pub fn with_seeding_alpha(self, alpha: f32) -> Self

Set seeding alpha (exponent for k-means++ probability weighting).

Standard k-means++ uses alpha=2.0 (D-squared weighting). Research (Bamas et al. 2023) suggests alpha > 2 (e.g. 4.0) can yield better final clustering cost.

Trait Implementations

impl<D: Clone + DistanceMetric> Clone for Kmeans<D>

fn clone(&self) -> Kmeans<D>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<D: DistanceMetric> Clustering for Kmeans<D>

fn fit_predict(&self, data: &[Vec<f32>]) -> Result<Vec<usize>>

Fit the model to data and return cluster assignments.

fn n_clusters(&self) -> usize

Get the number of clusters.

impl<D: Debug + DistanceMetric> Debug for Kmeans<D>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.