arroy 0.6.4

Annoy-inspired 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
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277

use std::fmt::Display;
use std::num::NonZeroUsize;

use roaring::RoaringBitmap;

use super::*;
use crate::distance::Cosine;
use crate::distances::{Euclidean, Manhattan};
use crate::{ItemId, Reader, Writer};

pub struct NnsRes(pub Option<Vec<(ItemId, f32)>>);

impl Display for NnsRes {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self.0 {
            Some(ref vec) => {
                for (id, dist) in vec {
                    writeln!(f, "id({id}): distance({dist})")?;
                }
                Ok(())
            }
            None => f.write_str("No results found"),
        }
    }
}

#[test]
fn open_unfinished_db() {
    let handle = create_database();
    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    writer.add_item(&mut wtxn, 0, &[0.0, 0.0]).unwrap();
    wtxn.commit().unwrap();

    let rtxn = handle.env.read_txn().unwrap();
    let ret = Reader::<Euclidean>::open(&rtxn, 0, handle.database).map(|_| ()).unwrap_err();
    insta::assert_snapshot!(ret, @"Metadata are missing on index 0, You must build your database before attempting to read it");
}

#[test]
fn open_db_with_wrong_dimension() {
    let handle = create_database();
    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    writer.add_item(&mut wtxn, 0, &[0.0, 0.0]).unwrap();

    writer.builder(&mut rng()).n_trees(1).build(&mut wtxn).unwrap();
    wtxn.commit().unwrap();

    let rtxn = handle.env.read_txn().unwrap();
    let reader = Reader::<Euclidean>::open(&rtxn, 0, handle.database).unwrap();
    let ret = reader.nns(5).by_vector(&rtxn, &[1.0, 2.0, 3.0]).unwrap_err();
    insta::assert_snapshot!(ret, @"Invalid vector dimensions. Got 3 but expected 2");
}

#[test]
fn open_db_with_wrong_distance() {
    let handle = create_database::<Euclidean>();
    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    writer.add_item(&mut wtxn, 0, &[0.0, 0.0]).unwrap();

    writer.builder(&mut rng()).n_trees(1).build(&mut wtxn).unwrap();
    wtxn.commit().unwrap();

    let rtxn = handle.env.read_txn().unwrap();
    let wrongly_typed_db = handle.database.remap_data_type::<NodeCodec<Manhattan>>();
    let err = Reader::open(&rtxn, 0, wrongly_typed_db).unwrap_err();
    insta::assert_debug_snapshot!(err, @r###"
    UnmatchingDistance {
        expected: "euclidean",
        received: "manhattan",
    }
    "###);
}

#[test]
fn search_in_db_with_a_single_vector() {
    // https://github.com/meilisearch/meilisearch/pull/4296
    let handle = create_database::<Cosine>();
    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 3);
    writer.add_item(&mut wtxn, 0, &[0.00397, 0.553, 0.0]).unwrap();

    writer.builder(&mut rng()).build(&mut wtxn).unwrap();
    wtxn.commit().unwrap();

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

    let ret = reader.nns(1).by_item(&rtxn, 0).unwrap();
    insta::assert_snapshot!(NnsRes(ret), @r###"
    id(0): distance(0)
    "###);
}

#[test]
fn two_dimension_on_a_line() {
    let handle = create_database();
    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    // We'll draw a simple line over the y as seen below
    // (0,0) # # # # # # ...
    for i in 0..100 {
        writer.add_item(&mut wtxn, i, &[i as f32, 0.0]).unwrap();
    }

    writer.builder(&mut rng()).n_trees(50).build(&mut wtxn).unwrap();
    wtxn.commit().unwrap();

    let rtxn = handle.env.read_txn().unwrap();
    let reader = Reader::<Euclidean>::open(&rtxn, 0, handle.database).unwrap();

    // if we can't look into enough nodes we find some random points
    let ret = reader.nns(5).search_k(NonZeroUsize::new(1).unwrap()).by_item(&rtxn, 1).unwrap();
    insta::assert_snapshot!(NnsRes(ret), @r###"
    id(48): distance(47)
    id(92): distance(91)
    "###);

    // if we can look into all the node there is no inifinite loop and it works
    let ret =
        reader.nns(5).search_k(NonZeroUsize::new(usize::MAX).unwrap()).by_item(&rtxn, 0).unwrap();
    insta::assert_snapshot!(NnsRes(ret), @r###"
    id(0): distance(0)
    id(1): distance(1)
    id(2): distance(2)
    id(3): distance(3)
    id(4): distance(4)
    "###);

    let ret = reader.nns(5).by_item(&rtxn, 0).unwrap();
    insta::assert_snapshot!(NnsRes(ret), @r###"
    id(0): distance(0)
    id(1): distance(1)
    id(2): distance(2)
    id(3): distance(3)
    id(4): distance(4)
    "###);
}

#[test]
fn two_dimension_on_a_column() {
    let handle = create_database();
    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    // We'll draw a simple line over the y as seen below
    // (0,0) # . . . . .
    // (0,1) # . . . . .
    // (0,2) # . . . . .
    // (0,3) # . . . . .
    // [...]
    for i in 0..100 {
        writer.add_item(&mut wtxn, i, &[0.0, i as f32]).unwrap();
    }

    writer.builder(&mut rng()).n_trees(50).build(&mut wtxn).unwrap();
    wtxn.commit().unwrap();

    let rtxn = handle.env.read_txn().unwrap();
    let reader = Reader::<Euclidean>::open(&rtxn, 0, handle.database).unwrap();
    let ret = reader.nns(5).by_item(&rtxn, 0).unwrap();

    insta::assert_snapshot!(NnsRes(ret), @r###"
    id(0): distance(0)
    id(1): distance(1)
    id(2): distance(2)
    id(3): distance(3)
    id(4): distance(4)
    "###);
}

#[test]
fn get_item_ids() {
    let handle = create_database();
    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    for i in 0..10 {
        writer.add_item(&mut wtxn, i, &[0.0, i as f32]).unwrap();
    }

    writer.builder(&mut rng()).n_trees(50).build(&mut wtxn).unwrap();

    let reader = Reader::<Euclidean>::open(&wtxn, 0, handle.database).unwrap();
    let ret = reader.item_ids();

    insta::assert_debug_snapshot!(ret, @"RoaringBitmap<[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]>");
}

#[test]
fn filtering() {
    let handle = create_database();
    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    // We'll draw a simple line over the y as seen below
    // (0,0) # . . . . .
    // (0,1) # . . . . .
    // (0,2) # . . . . .
    // (0,3) # . . . . .
    // [...]
    for i in 0..100 {
        writer.add_item(&mut wtxn, i, &[0.0, i as f32]).unwrap();
    }

    writer.builder(&mut rng()).n_trees(50).build(&mut wtxn).unwrap();
    wtxn.commit().unwrap();

    let rtxn = handle.env.read_txn().unwrap();
    let reader = Reader::<Euclidean>::open(&rtxn, 0, handle.database).unwrap();

    let ret = reader.nns(5).candidates(&RoaringBitmap::from_iter(0..2)).by_item(&rtxn, 0).unwrap();
    insta::assert_snapshot!(NnsRes(ret), @r###"
    id(0): distance(0)
    id(1): distance(1)
    "###);

    let ret =
        reader.nns(5).candidates(&RoaringBitmap::from_iter(98..1000)).by_item(&rtxn, 0).unwrap();
    insta::assert_snapshot!(NnsRes(ret), @r###"
    id(98): distance(98)
    id(99): distance(99)
    "###);
}

#[test]
fn search_in_empty_database() {
    // See https://github.com/meilisearch/arroy/issues/75
    let handle = create_database::<Euclidean>();

    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    writer.builder(&mut rng()).build(&mut wtxn).unwrap();
    wtxn.commit().unwrap();

    let rtxn = handle.env.read_txn().unwrap();
    let reader = Reader::open(&rtxn, 0, handle.database).unwrap();
    let ret = reader.nns(10).by_vector(&rtxn, &[0., 0.]).unwrap();
    insta::assert_debug_snapshot!(ret, @"[]");
}

#[test]
fn try_reading_in_a_non_built_database() {
    // See https://github.com/meilisearch/arroy/issues/74
    let handle = create_database::<Euclidean>();

    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    writer.add_item(&mut wtxn, 0, &[0.0, 0.0]).unwrap();
    // We don't build the database
    wtxn.commit().unwrap();

    let rtxn = handle.env.read_txn().unwrap();
    let error = Reader::open(&rtxn, 0, handle.database).unwrap_err();
    insta::assert_debug_snapshot!(error, @r###"
    MissingMetadata(
        0,
    )
    "###);
    drop(rtxn);

    // we build the database once to get valid metadata
    let mut wtxn = handle.env.write_txn().unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    writer.builder(&mut rng()).build(&mut wtxn).unwrap();
    let writer = Writer::new(handle.database, 0, 2);
    writer.del_item(&mut wtxn, 0).unwrap();
    // We don't build the database; this leaves the database in a corrupted state
    wtxn.commit().unwrap();

    let rtxn = handle.env.read_txn().unwrap();
    let error = Reader::open(&rtxn, 0, handle.database).unwrap_err();
    insta::assert_debug_snapshot!(error, @r###"
    NeedBuild(
        0,
    )
    "###);
}