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
use super::{
common::{VectorDistanceMetric, VectorRelationBinding},
definition::CuvsOperation,
};
/// Query-local IVF-Flat build and search parameters.
#[derive(Debug, Clone, PartialEq)]
pub struct VectorIvfFlatParams {
/// Dense rows indexed into inverted lists.
pub dataset: VectorRelationBinding,
/// Dense query rows matched against the dataset.
pub queries: VectorRelationBinding,
/// Number of neighbors returned per query row.
pub k: usize,
/// Distance or similarity used to rank neighbors.
pub metric: VectorDistanceMetric,
/// Number of inverted lists; must not exceed dataset row count.
pub n_lists: u32,
/// Lists searched per query; must not exceed `n_lists`.
pub n_probes: u32,
/// Maximum iterations for fitting list centers.
pub kmeans_n_iters: u32,
/// Fraction of dataset rows used to fit list centers, in `(0, 1]`.
pub kmeans_trainset_fraction: f64,
}
impl VectorIvfFlatParams {
/// Uses registry defaults for list training and search tuning.
///
/// The supplied bindings, neighbor count, metric, and list count are retained
/// without validation.
pub fn new(
dataset: VectorRelationBinding,
queries: VectorRelationBinding,
k: usize,
metric: VectorDistanceMetric,
n_lists: u32,
) -> Self {
let descriptor = CuvsOperation::IvfFlat.descriptor();
Self {
dataset,
queries,
k,
metric,
n_lists,
n_probes: descriptor
.option("n_probes")
.and_then(|option| option.kind.default_u64())
.expect("IVF-Flat n_probes default") as u32,
kmeans_n_iters: descriptor
.option("kmeans_n_iters")
.and_then(|option| option.kind.default_u64())
.expect("IVF-Flat kmeans_n_iters default") as u32,
kmeans_trainset_fraction: descriptor
.option("kmeans_trainset_fraction")
.and_then(|option| option.kind.default_f64())
.expect("IVF-Flat training fraction default"),
}
}
}