Struct cogset::Kmeans

pub struct Kmeans<T> {
    // some fields omitted
}

Clustering via the k-means algorithm (aka Lloyd's algorithm).

k-means clustering aims to partition n observations into k clusters in which each observation belongs to the cluster with the nearest mean, serving as a prototype of the cluster.^wikipedia

This is a heuristic, iterative approximation to the true optimal assignment. The parameters used to control the approximation can be set via KmeansBuilder.

Examples

use cogset::{Euclid, Kmeans};

let data = [Euclid([0.0, 0.0]),
            Euclid([1.0, 0.5]),
            Euclid([0.2, 0.2]),
            Euclid([0.3, 0.8]),
            Euclid([0.0, 1.0])];
let k = 3;

let kmeans = Kmeans::new(&data, k);

println!("{:?}", kmeans.clusters());

Methods

impl<T> Kmeans<T> where Euclid<T>: Point + Euclidean + Clone

fn new(data: &[Euclid<T>], k: usize) -> Kmeans<T>

Run k-means on data with the default settings.

fn clusters(&self) -> Vec<(Euclid<T>, Vec<usize>)>

Retrieve the means and the clusters themselves that this k-means instance computed.

The clusters are represented by vectors of indexes into the original data.

fn converged(&self) -> Result<usize, usize>

Return whether the algorithm converged, and how many steps that took.

Ok is returned if the algorithm did meet the tolerance criterion, and Err if it reached the iteration limit instead.