rlx-umap 0.2.18

Parametric UMAP on RLX — full fit/transform API and k-NN building blocks.
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

// RLX — versatile ML compiler + runtime.
// Copyright (C) 2026 Eugene Hauptmann, Nataliya Kosmyna.
// SPDX-License-Identifier: MIT OR Apache-2.0

//! Reference pairwise distance matrices (row-major `[n, n]`).

const EPS: f32 = 1e-12;

/// Full Euclidean distance matrix: `sqrt(max(0, ||x_i - x_j||^2))`.
pub fn euclidean_pairwise_reference(x: &[f32], n: usize, d: usize) -> Vec<f32> {
    let mut out = vec![0f32; n * n];
    let mut sq_norms = vec![0f32; n];
    for i in 0..n {
        let mut s = 0f32;
        for f in 0..d {
            let v = x[i * d + f];
            s += v * v;
        }
        sq_norms[i] = s;
    }
    for i in 0..n {
        for j in 0..n {
            let mut dot = 0f32;
            for f in 0..d {
                dot += x[i * d + f] * x[j * d + f];
            }
            let sq = (sq_norms[i] + sq_norms[j] - 2.0 * dot).max(0.0);
            out[i * n + j] = sq.sqrt();
        }
    }
    out
}

/// Cosine **distance** `1 - cos(θ)`, clamped to `[0, 2]`. Diagonal is `0`.
pub fn cosine_pairwise_reference(x: &[f32], n: usize, d: usize) -> Vec<f32> {
    let mut norms = vec![EPS; n];
    for i in 0..n {
        let mut s = 0f32;
        for f in 0..d {
            let v = x[i * d + f];
            s += v * v;
        }
        norms[i] = s.sqrt().max(EPS);
    }
    let mut out = vec![0f32; n * n];
    for i in 0..n {
        for j in 0..n {
            if i == j {
                out[i * n + j] = 0.0;
                continue;
            }
            let mut dot = 0f32;
            for f in 0..d {
                dot += x[i * d + f] * x[j * d + f];
            }
            let sim = (dot / (norms[i] * norms[j])).clamp(-1.0, 1.0);
            out[i * n + j] = (1.0 - sim).max(0.0);
        }
    }
    out
}