vsdb 13.2.0

A std-collection-like database
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
140
141
142
143

//! Scalar types and distance metrics for vector similarity search.

use serde::{Deserialize, Serialize, de};
use std::iter::Sum;
use std::ops::{Add, Div, Mul, Neg, Sub};

/// A floating-point scalar suitable for vector components.
///
/// Implemented for `f32` and `f64`.
pub trait Scalar:
    Copy
    + Default
    + PartialOrd
    + Add<Output = Self>
    + Sub<Output = Self>
    + Mul<Output = Self>
    + Div<Output = Self>
    + Neg<Output = Self>
    + Sum
    + Serialize
    + de::DeserializeOwned
    + Send
    + Sync
    + 'static
{
    fn sqrt(self) -> Self;
    fn epsilon() -> Self;
    fn one() -> Self;
    fn zero() -> Self;
    fn total_cmp(&self, other: &Self) -> std::cmp::Ordering;
}

impl Scalar for f32 {
    #[inline]
    fn sqrt(self) -> Self {
        f32::sqrt(self)
    }
    #[inline]
    fn epsilon() -> Self {
        f32::EPSILON
    }
    #[inline]
    fn one() -> Self {
        1.0
    }
    #[inline]
    fn zero() -> Self {
        0.0
    }
    #[inline]
    fn total_cmp(&self, other: &Self) -> std::cmp::Ordering {
        f32::total_cmp(self, other)
    }
}

impl Scalar for f64 {
    #[inline]
    fn sqrt(self) -> Self {
        f64::sqrt(self)
    }
    #[inline]
    fn epsilon() -> Self {
        f64::EPSILON
    }
    #[inline]
    fn one() -> Self {
        1.0
    }
    #[inline]
    fn zero() -> Self {
        0.0
    }
    #[inline]
    fn total_cmp(&self, other: &Self) -> std::cmp::Ordering {
        f64::total_cmp(self, other)
    }
}

/// Trait for computing distance between two vectors.
///
/// Lower distance means more similar.  Generic over [`Scalar`] so the
/// same metric works for both `f32` and `f64` vectors.
pub trait DistanceMetric<S: Scalar = f32>:
    Clone + Default + Serialize + de::DeserializeOwned + Send + Sync + 'static
{
    /// Compute the distance between two vectors of equal length.
    fn distance(a: &[S], b: &[S]) -> S;
}

/// Euclidean (L2) squared distance.
#[derive(Clone, Default, Debug, Serialize, Deserialize)]
pub struct L2;

impl<S: Scalar> DistanceMetric<S> for L2 {
    #[inline]
    fn distance(a: &[S], b: &[S]) -> S {
        debug_assert_eq!(a.len(), b.len());
        a.iter()
            .zip(b)
            .map(|(x, y)| {
                let d = *x - *y;
                d * d
            })
            .sum()
    }
}

/// Cosine distance: `1.0 - cosine_similarity`.
#[derive(Clone, Default, Debug, Serialize, Deserialize)]
pub struct Cosine;

impl<S: Scalar> DistanceMetric<S> for Cosine {
    #[inline]
    fn distance(a: &[S], b: &[S]) -> S {
        debug_assert_eq!(a.len(), b.len());
        let (mut dot, mut na, mut nb) = (S::zero(), S::zero(), S::zero());
        for (x, y) in a.iter().zip(b) {
            dot = dot + *x * *y;
            na = na + *x * *x;
            nb = nb + *y * *y;
        }
        let denom = na.sqrt() * nb.sqrt();
        if denom < S::epsilon() {
            S::one()
        } else {
            S::one() - dot / denom
        }
    }
}

/// Inner product distance: `-(a . b)`.
///
/// Negated so that higher inner product = lower distance.
#[derive(Clone, Default, Debug, Serialize, Deserialize)]
pub struct InnerProduct;

impl<S: Scalar> DistanceMetric<S> for InnerProduct {
    #[inline]
    fn distance(a: &[S], b: &[S]) -> S {
        debug_assert_eq!(a.len(), b.len());
        -(a.iter().zip(b).map(|(x, y)| *x * *y).sum::<S>())
    }
}