zvec-bindings 0.4.1

Idiomatic Rust bindings for zvec vector database
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

use std::cmp::Ordering;
use std::collections::HashMap;

use crate::types::MetricType;

pub struct RrfReRanker {
    topn: usize,
    rank_constant: i32,
}

impl RrfReRanker {
    pub fn new(topn: usize) -> Self {
        Self {
            topn,
            rank_constant: 60,
        }
    }

    pub fn with_rank_constant(mut self, rank_constant: i32) -> Self {
        self.rank_constant = rank_constant;
        self
    }

    pub fn topn(&self) -> usize {
        self.topn
    }

    pub fn rank_constant(&self) -> i32 {
        self.rank_constant
    }

    fn rrf_score(&self, rank: usize) -> f64 {
        1.0 / (self.rank_constant as f64 + rank as f64 + 1.0)
    }

    pub fn rerank<K: AsRef<str>>(
        &self,
        query_results: &HashMap<K, Vec<(String, f32)>>,
    ) -> Vec<(String, f32)> {
        let mut rrf_scores: HashMap<String, f64> = HashMap::new();

        for docs in query_results.values() {
            for (rank, (doc_id, _)) in docs.iter().enumerate() {
                let rrf_score = self.rrf_score(rank);
                *rrf_scores.entry(doc_id.clone()).or_insert(0.0) += rrf_score;
            }
        }

        let mut scored: Vec<_> = rrf_scores.into_iter().collect();
        scored.sort_by(|a, b| b.1.partial_cmp(&a.1).unwrap_or(Ordering::Equal));
        scored.truncate(self.topn);

        scored
            .into_iter()
            .map(|(id, score)| (id, score as f32))
            .collect()
    }
}

pub struct WeightedReRanker {
    topn: usize,
    metric: MetricType,
    weights: HashMap<String, f64>,
}

impl WeightedReRanker {
    pub fn new(topn: usize, metric: MetricType) -> Self {
        Self {
            topn,
            metric,
            weights: HashMap::new(),
        }
    }

    pub fn with_weight(mut self, field: impl Into<String>, weight: f64) -> Self {
        self.weights.insert(field.into(), weight);
        self
    }

    pub fn with_weights(mut self, weights: HashMap<String, f64>) -> Self {
        self.weights = weights;
        self
    }

    pub fn topn(&self) -> usize {
        self.topn
    }

    pub fn metric(&self) -> MetricType {
        self.metric
    }

    fn normalize_score(&self, score: f32) -> f64 {
        match self.metric {
            MetricType::L2 => 1.0 - 2.0 * (score as f64).atan() / std::f64::consts::PI,
            MetricType::Ip => 0.5 + (score as f64).atan() / std::f64::consts::PI,
            MetricType::Cosine => 1.0 - score as f64 / 2.0,
            _ => score as f64,
        }
    }

    pub fn rerank<K: AsRef<str>>(
        &self,
        query_results: &HashMap<K, Vec<(String, f32)>>,
    ) -> Vec<(String, f32)> {
        let mut weighted_scores: HashMap<String, f64> = HashMap::new();

        for (vector_name, docs) in query_results.iter() {
            let weight = self
                .weights
                .get(vector_name.as_ref())
                .copied()
                .unwrap_or(1.0);
            for (doc_id, score) in docs.iter() {
                let normalized = self.normalize_score(*score);
                let weighted = normalized * weight;
                *weighted_scores.entry(doc_id.clone()).or_insert(0.0) += weighted;
            }
        }

        let mut scored: Vec<_> = weighted_scores.into_iter().collect();
        scored.sort_by(|a, b| b.1.partial_cmp(&a.1).unwrap_or(Ordering::Equal));
        scored.truncate(self.topn);

        scored
            .into_iter()
            .map(|(id, score)| (id, score as f32))
            .collect()
    }
}