hannoy 0.1.3

HNSW Approximate Nearest Neighbors in Rust, based on LMDB and optimized for memory usage
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
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170

#[cfg(not(windows))]
use proptest::prelude::*;
use rand::rngs::StdRng;
use rand::seq::SliceRandom;
use rand::{thread_rng, Rng, SeedableRng};
use roaring::RoaringBitmap;

use crate::distance::{BinaryQuantizedCosine, Cosine};
use crate::tests::{create_database, create_database_indices_with_items, rng, DatabaseHandle};
use crate::{Reader, Writer};

const M: usize = 16;
const M0: usize = 32;

// Minimal reproducer for issue #78
// <https://github.com/nnethercott/hannoy/issues/78>
#[test]
fn quantized_iter_has_right_dimensions() {
    let DatabaseHandle { env, database, tempdir: _ } = create_database::<BinaryQuantizedCosine>();
    let mut wtxn = env.write_txn().unwrap();
    // use a prime number of dims
    const DIM: usize = 1063;
    let writer = Writer::new(database, 0, DIM);

    let mut rng = StdRng::seed_from_u64(42);

    let mut vec = [0f32; DIM];
    rng.fill(&mut vec);
    writer.add_item(&mut wtxn, 0, &vec).unwrap();
    writer.builder(&mut rng).build::<M, M0>(&mut wtxn).unwrap();
    wtxn.commit().unwrap();

    let rtxn = env.read_txn().unwrap();
    let reader = Reader::open(&rtxn, 0, database).unwrap();
    let mut cursor = reader.iter(&rtxn).unwrap();
    let (_, new_vec) = cursor.next().unwrap().unwrap();

    assert!(new_vec.len() == DIM);
}

#[test]
fn search_on_candidates_has_right_num() {
    const DIM: usize = 768;
    let db_indexes = 1..5;

    let DatabaseHandle { env, database, tempdir: _ } =
        create_database_indices_with_items::<Cosine, DIM, M, M0, _>(
            db_indexes.clone(),
            1000,
            &mut rng(),
        );

    let rtxn = env.read_txn().unwrap();
    let mut rng = rng();
    let mut shuffled_indices = Vec::from_iter(db_indexes);
    shuffled_indices.shuffle(&mut rng);

    for index in shuffled_indices {
        let reader = crate::Reader::<Cosine>::open(&rtxn, index, database).unwrap();

        // search with 10 candidates
        let mut query = [f32::default(); DIM];
        rng.fill(&mut query);

        let c: [u32; 10] = std::array::from_fn(|_| thread_rng().gen::<u32>() % 1000);
        let candidates = RoaringBitmap::from_iter(c);
        let _found = reader.nns(10).candidates(&candidates).by_vector(&rtxn, &query).unwrap();
        let found = _found.into_nns();
        assert_eq!(&RoaringBitmap::from_iter(found.into_iter().map(|(i, _)| i)), &candidates);

        // search with 1 candidate
        let c: [u32; 1] = std::array::from_fn(|_| thread_rng().gen::<u32>() % 1000);
        let candidates = RoaringBitmap::from_iter(c);
        let _found = reader.nns(1).candidates(&candidates).by_vector(&rtxn, &query).unwrap();
        let found = _found.into_nns();
        assert_eq!(&RoaringBitmap::from_iter(found.into_iter().map(|(i, _)| i)), &candidates);
    }
}

// ci takes too long on windows
#[cfg(not(windows))]
fn all_items_are_reachable<const M: usize, const M0: usize>(n: usize) {
    const DIM: usize = 768;
    let mut rng = rng();

    let DatabaseHandle { env, database, tempdir: _ } =
        create_database_indices_with_items::<Cosine, DIM, M, M0, _>(0..1, n, &mut rng);
    let rtxn = env.read_txn().unwrap();

    // Check that all items were written correctly
    let reader = crate::Reader::<Cosine>::open(&rtxn, 0, database).unwrap();
    assert_eq!(reader.item_ids().len(), n as u64);
    assert!((0..n as u32).all(|i| reader.contains_item(&rtxn, i).unwrap()));

    let _found = reader.nns(n).ef_search(n).by_vector(&rtxn, &[0.0; DIM]).unwrap();
    let found = _found.into_nns();
    assert_eq!(&RoaringBitmap::from_iter(found.into_iter().map(|(id, _)| id)), reader.item_ids())
}

// ci takes too long on windows
#[cfg(not(windows))]
proptest! {
    #![proptest_config(ProptestConfig {
            cases: 10,.. ProptestConfig::default()
    })]

    #[test]
    fn all_items_are_reachable_3_3(n in 1..10000usize){
        all_items_are_reachable::<6,6>(n);
    }
}

#[test]
fn search_by_item_does_not_contain_item() {
    const DIM: usize = 768;
    let mut rng = rng();

    let DatabaseHandle { env, database, tempdir: _ } =
        create_database_indices_with_items::<Cosine, DIM, M, M0, _>(0..1, 100, &mut rng);
    let rtxn = env.read_txn().unwrap();

    let reader = crate::Reader::<Cosine>::open(&rtxn, 0, database).unwrap();

    let found = reader.nns(10).by_item(&rtxn, 0).unwrap().unwrap().into_nns();
    assert!(found.len() == 10);
    assert!(!found.contains(&(0, 0.0)))
}

#[test]
fn search_by_item_returns_none_if_not_exists() {
    const DIM: usize = 768;
    let mut rng = rng();

    let DatabaseHandle { env, database, tempdir: _ } =
        create_database_indices_with_items::<Cosine, DIM, M, M0, _>(0..1, 100, &mut rng);
    let rtxn = env.read_txn().unwrap();

    let reader = crate::Reader::<Cosine>::open(&rtxn, 0, database).unwrap();

    // use an item id that does not exist
    let found = reader.nns(10).by_item(&rtxn, 101).unwrap();
    assert!(found.is_none());
}

#[test]
fn search_cancellation_works() {
    const DIM: usize = 768;
    let mut rng = rng();

    let DatabaseHandle { env, database, tempdir: _ } =
        create_database_indices_with_items::<Cosine, DIM, M, M0, _>(0..1, 100, &mut rng);
    let rtxn = env.read_txn().unwrap();

    let reader = crate::Reader::<Cosine>::open(&rtxn, 0, database).unwrap();

    // use an item id that does not exist
    let query: [f32; DIM] = std::array::from_fn(|_| rng.gen());

    // by vector
    let searched = reader.nns(10).by_vector_with_cancellation(&rtxn, &query, || false).unwrap();
    assert!(!searched.did_cancel());
    let searched = reader.nns(10).by_vector_with_cancellation(&rtxn, &query, || true).unwrap();
    assert!(searched.did_cancel());

    // by item
    let searched = reader.nns(10).by_item_with_cancellation(&rtxn, 0, || false).unwrap().unwrap();
    assert!(!searched.did_cancel());
    let searched = reader.nns(10).by_item_with_cancellation(&rtxn, 0, || true).unwrap().unwrap();
    assert!(searched.did_cancel());
}