kannolo 0.1.6

kANNolo is designed for easy prototyping of ANN Search algorithms while ensuring high effectiveness and efficiency over both dense and sparse vectors.
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
278
279
280
281
282
283
284
285
286
287
288

use std::{fmt::Debug, time::Instant};

use clap::Parser;
use kannolo::plain_quantizer::PlainQuantizer;
use kannolo::pq::ProductQuantizer;
use rand::{rngs::StdRng, seq::IteratorRandom, SeedableRng};
use std::process;

use kannolo::{
    hnsw::graph_index::GraphIndex, hnsw_utils::config_hnsw::ConfigHnsw, Dataset, DistanceType,
    IndexSerializer,
};
use kannolo::{read_numpy_f32_flatten_2d, DArray1, DenseDataset};

#[derive(Parser, Debug)]
#[clap(author, version, about, long_about = None)]
struct Args {
    /// The path of the dataset seismic file. Only one between data_file and index_file must be provided.
    /// If both are provided, the index_file will be used.
    #[clap(short, long, value_parser)]
    data_file: String,

    /// The output file where to save the index.
    #[clap(short, long, value_parser)]
    output_file: String,

    /// The number of neihbors per node.
    #[clap(long, value_parser)]
    #[arg(default_value_t = 16)]
    m: usize,

    /// The size of the candidate pool at construction time.
    #[clap(long, value_parser)]
    #[arg(default_value_t = 40)]
    efc: usize,

    /// The number of subspaces of Product Quantization.
    #[clap(long, value_parser)]
    #[arg(default_value_t = 16)]
    m_pq: usize,

    /// The number of bits of each subspace of Product Quantization.
    #[clap(long, value_parser)]
    #[arg(default_value_t = 8)]
    nbits: usize,

    /// The type of distance to use. Either 'l2' (Euclidean) or 'ip' (Inner product).
    #[clap(long, value_parser)]
    #[arg(default_value_t = String::from("ip"))]
    metric: String,

    /// The size of the sample of the dataset used for the training of Product Quantization.
    #[clap(long, value_parser)]
    #[arg(default_value_t = 100_000)]
    sample_size: usize,
}

fn main() {
    // Parse command line arguments
    let args: Args = Args::parse();

    let data_path = args.data_file;

    let num_neighbors = args.m;
    let ef_construction = args.efc;
    let m_pq = args.m_pq;
    let sample_size = args.sample_size;
    let nbits = args.nbits;

    println!("Building Index with M: {num_neighbors}, ef_construction: {ef_construction}");

    let distance = match args.metric.as_str() {
        "l2" => DistanceType::Euclidean,
        "ip" => DistanceType::DotProduct,
        _ => {
            eprintln!("Error: Invalid distance type. Choose between 'l2' and 'ip'.");
            process::exit(1);
        }
    };

    let num_threads_construction = rayon::current_num_threads();

    // Set parameters for the HNSW index
    let config = ConfigHnsw::new()
        .num_neighbors(num_neighbors)
        .ef_construction(ef_construction)
        .build();

    let (docs_vec, d) = read_numpy_f32_flatten_2d(data_path.to_string());
    let dataset = DenseDataset::from_vec(docs_vec, d, PlainQuantizer::<f32>::new(d, distance));

    match m_pq {
        8 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<8>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        16 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<16>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        32 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<32>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        48 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<48>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        64 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<64>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        96 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<96>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        128 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<128>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        192 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<192>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        256 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<256>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        384 => {
            let mut rng = StdRng::seed_from_u64(523);
            let mut training_vec: Vec<f32> = Vec::new();
            for vec in dataset.iter().choose_multiple(&mut rng, sample_size) {
                training_vec.extend(vec.values_as_slice());
            }
            let training_dataset = DenseDataset::from_vec(
                training_vec,
                dataset.dim(),
                PlainQuantizer::<f32>::new(dataset.dim(), distance),
            );
            let quantizer = ProductQuantizer::<384>::train(&training_dataset, nbits, distance);
            let start_time = Instant::now();
            let index =
                GraphIndex::from_dataset(&dataset, &config, quantizer, num_threads_construction);
            let duration = start_time.elapsed();
            println!("Time to build {} (before serializing)", duration.as_secs());
            let _ = IndexSerializer::save_index(&args.output_file, &index);
        }
        _ => {
            eprintln!("Error: Invalid number of subspaces for Product Quantization. Available values are 8, 16, 32, 64, 96, 128, 192, 256, 384.");
            process::exit(1);
        }
    };
}