recern-vector 0.0.2

A single-file, embedded, inspectable vector 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

//! Builds an index over synthetic clustered vectors and reports build speed,
//! recall and latency.
//!
//! cargo run --release --example bench -- [records] [dim] [noise]
//!
//! `noise` is the per-dimension spread around cluster centers (default 1.0);
//! higher values make neighbors harder to tell apart.

use std::time::Instant;

use recern_vector::{CollectionConfig, Database, Metric, RecallOptions};

/// Clustered data resembles real embeddings far better than uniform noise.
struct Clusters {
    state: u64,
    noise: f32,
    centers: Vec<Vec<f32>>,
}

impl Clusters {
    fn new(count: usize, dim: usize, noise: f32, seed: u64) -> Self {
        let mut this = Self {
            state: seed,
            noise,
            centers: Vec::new(),
        };
        this.centers = (0..count)
            .map(|_| (0..dim).map(|_| this.gaussian()).collect())
            .collect();
        this
    }

    fn uniform(&mut self) -> f32 {
        self.state ^= self.state << 13;
        self.state ^= self.state >> 7;
        self.state ^= self.state << 17;
        ((self.state >> 40) as f32 + 1.0) / (1u64 << 24) as f32
    }

    fn gaussian(&mut self) -> f32 {
        let (u, v) = (self.uniform(), self.uniform());
        (-2.0 * u.ln()).sqrt() * (std::f32::consts::TAU * v).cos()
    }

    fn sample(&mut self) -> Vec<f32> {
        let center = (self.uniform() * self.centers.len() as f32) as usize % self.centers.len();
        let center = self.centers[center].clone();
        center
            .iter()
            .map(|c| c + self.noise * self.gaussian())
            .collect()
    }
}

fn main() -> recern_vector::Result<()> {
    let args: Vec<String> = std::env::args().skip(1).collect();
    let arg = |i: usize| {
        args.get(i)
            .map(|a| a.parse::<f64>().expect("numeric argument"))
    };
    let records = arg(0).map_or(50_000, |v| v as usize);
    let dim = arg(1).map_or(128, |v| v as usize);
    let noise = arg(2).map_or(1.0, |v| v as f32);
    let path = std::env::temp_dir().join("recern-vector-bench.rvec");
    let _ = std::fs::remove_file(&path);

    let mut data = Clusters::new(100, dim, noise, 0x5EED);
    let vectors: Vec<Vec<f32>> = (0..records).map(|_| data.sample()).collect();

    let mut db = Database::create(&path)?;
    let collection = db.create_collection("bench", CollectionConfig::new(dim, Metric::Cosine))?;
    let start = Instant::now();
    for (i, v) in vectors.iter().enumerate() {
        collection.upsert(&i.to_string(), v, None)?;
    }
    let build = start.elapsed();

    let start = Instant::now();
    db.save()?;
    let save = start.elapsed();
    let start = Instant::now();
    let db = Database::open(&path)?;
    let open = start.elapsed();
    let collection = db.collection("bench")?;
    let stats = collection.stats();

    println!("records {records} · dim {dim} · noise {noise} · cosine · m=16 ef_construction=200");
    println!(
        "build   {:.2} s ({:.0} inserts/s)",
        build.as_secs_f64(),
        records as f64 / build.as_secs_f64()
    );
    println!(
        "file    {:.1} MB · save {:.0} ms · open {:.0} ms",
        std::fs::metadata(&path)?.len() as f64 / 1_048_576.0,
        save.as_secs_f64() * 1e3,
        open.as_secs_f64() * 1e3
    );
    println!(
        "graph   layers {:?} · avg degree L0 {:.1} · unreachable {}",
        stats.nodes_per_layer, stats.avg_degree_layer0, stats.unreachable
    );

    let report = collection.estimate_recall(&RecallOptions {
        sample: 200,
        k: 10,
        ef_values: vec![10, 16, 32, 64, 128, 256],
        seed: 7,
    })?;
    println!();
    println!(
        "exact scan p50 {:.3} ms",
        report.exact_p50.as_secs_f64() * 1e3
    );
    println!(
        "{:>6}  {:>9}  {:>10}  {:>10}  {:>9}",
        "ef", "recall@10", "p50 ms", "p95 ms", "speedup"
    );
    for p in &report.points {
        println!(
            "{:>6}  {:>9.3}  {:>10.3}  {:>10.3}  {:>8.0}x",
            p.ef,
            p.recall,
            p.p50.as_secs_f64() * 1e3,
            p.p95.as_secs_f64() * 1e3,
            report.exact_p50.as_secs_f64() / p.p50.as_secs_f64()
        );
    }

    let _ = std::fs::remove_file(&path);
    Ok(())
}