hdbscan 0.12.0

HDBSCAN clustering in pure Rust. A huge improvement on DBSCAN, capable of identifying clusters of varying densities.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139

#![cfg(feature = "parallel")]
use super::{NnAlgorithm, BRUTE_FORCE_N_SAMPLES_LIMIT};
use crate::distance::{get_dist_func, DistanceMetric};
use crate::HdbscanHyperParams;
use num_traits::Float;
use rayon::prelude::*;

const BRUTE_CHUNK_DATASET_THRESHOLD: usize = 10_000;

pub(crate) struct CoreDistanceCalculatorPar<'a, T> {
    data: &'a [Vec<T>],
    nn_algo: NnAlgorithm,
    dist_metric: DistanceMetric,
    k: usize,
}

impl<'a, T: Float + Send + Sync> CoreDistanceCalculatorPar<'a, T> {
    pub(crate) fn new(data: &'a [Vec<T>], hp: &'a HdbscanHyperParams) -> Self {
        Self {
            data,
            nn_algo: hp.nn_algo.clone(),
            dist_metric: hp.dist_metric,
            k: hp.min_samples,
        }
    }

    pub(crate) fn calc_core_distances(&self) -> Vec<T> {
        let n_samples = self.data.len();
        match (&self.nn_algo, n_samples, &self.dist_metric) {
            (_, _, DistanceMetric::Precalculated) => {
                get_core_distances_from_matrix(self.data, self.k)
            }
            (NnAlgorithm::Auto, usize::MIN..=BRUTE_FORCE_N_SAMPLES_LIMIT, _) => {
                BruteForce::calc_core_distances_direct(self.data, self.k, self.dist_metric)
            }
            (NnAlgorithm::Auto, _, _) => {
                KdTree::calc_core_distances(self.data, self.k, self.dist_metric)
            }
            (NnAlgorithm::BruteForce, usize::MIN..=BRUTE_CHUNK_DATASET_THRESHOLD, _) => {
                BruteForce::calc_core_distances_direct(self.data, self.k, self.dist_metric)
            }
            (NnAlgorithm::BruteForce, _, _) => {
                BruteForce::calc_core_distances_chunked(self.data, self.k, self.dist_metric)
            }
            (NnAlgorithm::KdTree, _, _) => {
                KdTree::calc_core_distances(self.data, self.k, self.dist_metric)
            }
        }
    }
}

pub(crate) fn get_core_distances_from_matrix<T>(dist_matrix: &[Vec<T>], k: usize) -> Vec<T>
where
    T: Float + Send + Sync,
{
    dist_matrix
        .par_iter()
        .map(|distances| {
            let mut dist = distances.clone();
            dist.sort_by(|a, b| a.partial_cmp(b).expect("Invalid float"));
            dist[k - 1]
        })
        .collect()
}

pub(crate) struct BruteForce;

impl BruteForce {
    fn calc_core_distances_direct<T: Float + Send + Sync>(
        data: &[Vec<T>],
        k: usize,
        dist_metric: DistanceMetric,
    ) -> Vec<T> {
        let dist_func = get_dist_func(&dist_metric);

        data.par_iter()
            .map(|point| {
                let mut distances: Vec<T> = data
                    .par_iter()
                    .map(|other| dist_func(point, other))
                    .collect();
                distances.par_sort_by(|a, b| a.partial_cmp(b).expect("Invalid float"));
                distances[k - 1]
            })
            .collect()
    }

    fn calc_core_distances_chunked<T: Float + Send + Sync>(
        data: &[Vec<T>],
        k: usize,
        dist_metric: DistanceMetric,
    ) -> Vec<T> {
        let chunk_size = (data.len() / rayon::current_num_threads()).max(100);
        let dist_func = get_dist_func(&dist_metric);

        data.par_chunks(chunk_size)
            .flat_map(|chunk| {
                chunk
                    .iter()
                    .map(|point| {
                        let mut distances: Vec<T> =
                            data.iter().map(|other| dist_func(point, other)).collect();
                        distances.sort_by(|a, b| a.partial_cmp(b).expect("Invalid float"));
                        distances[k - 1]
                    })
                    .collect::<Vec<_>>()
            })
            .collect()
    }
}

pub(crate) struct KdTree;

impl KdTree {
    fn calc_core_distances<T: Float + Send + Sync>(
        data: &[Vec<T>],
        k: usize,
        dist_metric: DistanceMetric,
    ) -> Vec<T> {
        let mut tree: kdtree::KdTree<T, usize, &Vec<T>> = kdtree::KdTree::new(data[0].len());
        data.iter()
            .enumerate()
            .for_each(|(n, datapoint)| tree.add(datapoint, n).expect("Failed to add to KdTree"));

        let dist_func = get_dist_func(&dist_metric);
        data.par_iter()
            .map(|datapoint| {
                let result = tree
                    .nearest(datapoint, k, &dist_func)
                    .expect("Failed to find neighbours");
                result
                    .into_iter()
                    .map(|(dist, _idx)| dist)
                    .last()
                    .expect("Failed to find neighbours")
            })
            .collect()
    }
}