Struct hdbscan::Hdbscan

source ·

pub struct Hdbscan<'a, T> { /* private fields */ }

Expand description

The HDBSCAN clustering algorithm in Rust. Generic over floating point numeric types.

Implementations§

source §

impl<'a, T: Float> Hdbscan<'a, T>

source

pub fn new(data: &'a Vec<Vec<T>>, hyper_params: HdbscanHyperParams) -> Self

Creates an instance of HDBSCAN clustering model using a custom hyper parameter configuration.

§Parameters

data - a reference to the data to cluster, a collection of vectors of floating points numbers. The vectors must all be of the same dimensionality and contain no infinite values.
config - the hyper parameter configuration.

§Returns

The HDBSCAN model instance.

§Examples

use hdbscan::{DistanceMetric, Hdbscan, HdbscanHyperParams, NnAlgorithm};

let data: Vec<Vec<f32>> = vec![
   vec![1.3, 1.1],
   vec![1.3, 1.2],
   vec![1.0, 1.1],
   vec![1.2, 1.2],
   vec![0.9, 1.0],
   vec![0.9, 1.0],
   vec![3.7, 4.0],
   vec![3.9, 3.9],
];
let config = HdbscanHyperParams::builder()
   .min_cluster_size(3)
   .min_samples(2)
   .dist_metric(DistanceMetric::Manhattan)
   .nn_algorithm(NnAlgorithm::BruteForce)
   .build();
let clusterer = Hdbscan::new(&data, config);

source

pub fn default(data: &'a Vec<Vec<T>>) -> Hdbscan<'_, T>

Creates an instance of HDBSCAN clustering model using the default hyper parameters.

§Parameters

data - a reference to the data to cluster, a collection of vectors of floating points numbers. The vectors must all be of the same dimensionality and contain no infinite values.

§Returns

The HDBSCAN model instance.

§Examples

use hdbscan::Hdbscan;

let data: Vec<Vec<f32>> = vec![
   vec![1.3, 1.1],
   vec![1.3, 1.2],
   vec![1.0, 1.1],
   vec![1.2, 1.2],
   vec![0.9, 1.0],
   vec![0.9, 1.0],
   vec![3.7, 4.0],
   vec![3.9, 3.9],
];
let clusterer = Hdbscan::default(&data);

source

pub fn cluster(&self) -> Result<Vec<i32>, HdbscanError>

Performs clustering on the list of vectors passed to the constructor.

§Returns

A result that, if successful, contains a list of cluster labels, with a length equal to the numbe of samples passed to the constructor. Positive integers mean a data point belongs to a cluster of that label. -1 labels mean that a data point is noise and does not belong to any cluster. An Error will be returned if the dimensionality of the input vectors are mismatched, if any vector contains non-finite coordinates, or if the passed data set is empty.

§Examples

use std::collections::HashSet;
use hdbscan::Hdbscan;

let data: Vec<Vec<f32>> = vec![
   vec![1.5, 2.2],
   vec![1.0, 1.1],
   vec![1.2, 1.4],
   vec![0.8, 1.0],
   vec![1.1, 1.0],
   vec![3.7, 4.0],
   vec![3.9, 3.9],
   vec![3.6, 4.1],
   vec![3.8, 3.9],
   vec![4.0, 4.1],
   vec![10.0, 10.0],
];
let clusterer = Hdbscan::default(&data);
let labels = clusterer.cluster().unwrap();
//First five points form one cluster
assert_eq!(1, labels[..5].iter().collect::<HashSet<_>>().len());
// Next five points are a second cluster
assert_eq!(1, labels[5..10].iter().collect::<HashSet<_>>().len());
// The final point is noise
assert_eq!(-1, labels[10]);

source

pub fn calc_centers( &self, center: Center, labels: &[i32], ) -> Result<Vec<Vec<T>>, HdbscanError>

Calculates the centers of the clusters just calculate.

§Parameters

center - the type of center to calculate. Currently only centroid (the element wise mean of all the data points in a cluster) is supported.
labels - a reference to the labels calculated by a call to Hdbscan::cluster.

§Returns

A vector of the cluster centers, of shape num clusters by num dimensions/features.

§Panics

If the labels are of different length to the data passed to the Hdbscan constructor

§Examples

use hdbscan::{Center, Hdbscan};

let data: Vec<Vec<f32>> = vec![
   vec![1.5, 2.2],
   vec![1.0, 1.1],
   vec![1.2, 1.4],
   vec![0.8, 1.0],
   vec![1.1, 1.0],
   vec![3.7, 4.0],
   vec![3.9, 3.9],
   vec![3.6, 4.1],
   vec![3.8, 3.9],
   vec![4.0, 4.1],
   vec![10.0, 10.0],
];
let clusterer = Hdbscan::default(&data);
let labels = clusterer.cluster().unwrap();
let centroids = clusterer.calc_centers(Center::Centroid, &labels).unwrap();
assert_eq!(2, centroids.len());
assert!(centroids.contains(&vec![3.8, 4.0]) && centroids.contains(&vec![1.12, 1.34]));