vectus 0.1.38

A vector database implemented in Rust for learning purposes.
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

use ndarray::Array1;

#[derive(Default)]
pub enum Metric {
    #[default]
    Cosine,
    Euclidean,
    Manhattan,
    Jaccard,
}

impl Metric {
    pub(crate) fn distance(&self, emb1: &Array1<f32>, emb2: &Array1<f32>) -> f32 {
        match self {
            Metric::Euclidean => self.euclidean_distance(emb1, emb2),
            Metric::Cosine => self.cosine_distance(emb1, emb2),
            Metric::Manhattan => self.manhattan_distance(emb1, emb2),
            Metric::Jaccard => self.jaccard_distance(emb1, emb2),
        }
    }

    fn euclidean_distance(&self, emb1: &Array1<f32>, emb2: &Array1<f32>) -> f32 {
        let diff = emb1 - emb2;
        diff.dot(&diff).sqrt()
    }

    fn manhattan_distance(&self, emb1: &Array1<f32>, emb2: &Array1<f32>) -> f32 {
        emb1.iter()
            .zip(emb2.iter())
            .map(|(a, b)| (a - b).abs())
            .sum()
    }

    fn jaccard_distance(&self, emb1: &Array1<f32>, emb2: &Array1<f32>) -> f32 {
        let intersection = emb1.iter().zip(emb2.iter()).filter(|(a, b)| a == b).count();
        let union = emb1.len() + emb2.len() - intersection;
        1.0 - (intersection as f32 / union as f32)
    }

    fn cosine_distance(&self, emb1: &Array1<f32>, emb2: &Array1<f32>) -> f32 {
        let dot_product = emb1.dot(emb2);
        let norm1 = emb1.dot(emb1).sqrt();
        let norm2 = emb2.dot(emb2).sqrt();
        1.0 - dot_product / (norm1 * norm2)
    }
}