nodedb_vector/distance/
scalar.rs

1//! Scalar (non-SIMD) distance metric implementations.
2//!
3//! Re-exports scalar implementations from `nodedb-types`.
4//! Works on all targets: native, WASM, iOS, Android.
5
6pub use nodedb_types::vector_distance::*;
7
8/// Compute distance between two vectors using scalar implementations.
9pub fn scalar_distance(a: &[f32], b: &[f32], metric: super::DistanceMetric) -> f32 {
10    use super::DistanceMetric::*;
11    match metric {
12        L2 => l2_squared(a, b),
13        Cosine => cosine_distance(a, b),
14        InnerProduct => neg_inner_product(a, b),
15        Manhattan => manhattan(a, b),
16        Chebyshev => chebyshev(a, b),
17        Hamming => hamming_f32(a, b),
18        Jaccard => jaccard(a, b),
19        Pearson => pearson(a, b),
20    }
21}
22
23#[cfg(test)]
24mod tests {
25    use super::*;
26
27    #[test]
28    fn l2_identical_is_zero() {
29        let v = [1.0, 2.0, 3.0];
30        assert_eq!(l2_squared(&v, &v), 0.0);
31    }
32
33    #[test]
34    fn l2_known_distance() {
35        let a = [0.0, 0.0];
36        let b = [3.0, 4.0];
37        assert_eq!(l2_squared(&a, &b), 25.0);
38    }
39
40    #[test]
41    fn cosine_identical_is_zero() {
42        let v = [1.0, 2.0, 3.0];
43        assert!(cosine_distance(&v, &v) < 1e-6);
44    }
45
46    #[test]
47    fn cosine_orthogonal_is_one() {
48        let a = [1.0, 0.0];
49        let b = [0.0, 1.0];
50        assert!((cosine_distance(&a, &b) - 1.0).abs() < 1e-6);
51    }
52
53    #[test]
54    fn neg_ip_basic() {
55        let a = [1.0, 2.0];
56        let b = [3.0, 4.0];
57        assert_eq!(neg_inner_product(&a, &b), -11.0);
58    }
59
60    #[test]
61    fn manhattan_basic() {
62        let a = [1.0, 2.0, 3.0];
63        let b = [4.0, 6.0, 3.0];
64        assert_eq!(manhattan(&a, &b), 7.0);
65    }
66
67    #[test]
68    fn chebyshev_basic() {
69        let a = [1.0, 2.0, 3.0];
70        let b = [4.0, 6.0, 3.0];
71        assert_eq!(chebyshev(&a, &b), 4.0);
72    }
73
74    #[test]
75    fn hamming_basic() {
76        let a = [1.0, 0.0, 1.0, 0.0];
77        let b = [1.0, 1.0, 0.0, 0.0];
78        assert_eq!(hamming_f32(&a, &b), 2.0);
79    }
80
81    #[test]
82    fn jaccard_basic() {
83        let a = [1.0, 0.0, 1.0, 0.0];
84        let b = [1.0, 1.0, 0.0, 0.0];
85        let j = jaccard(&a, &b);
86        assert!((j - (1.0 - 1.0 / 3.0)).abs() < 1e-6);
87    }
88
89    #[test]
90    fn pearson_identical_is_zero() {
91        let v = [1.0, 2.0, 3.0, 4.0, 5.0];
92        assert!(pearson(&v, &v) < 1e-6);
93    }
94
95    #[test]
96    fn pearson_opposite_is_high() {
97        let a = [1.0, 2.0, 3.0, 4.0, 5.0];
98        let b = [5.0, 4.0, 3.0, 2.0, 1.0];
99        assert!(pearson(&a, &b) > 1.5);
100    }
101}
nodedb_vector/distance/scalar.rs

nodedb_vector/distance/
scalar.rs
