fornix 0.3.0

Knowledge storage, retrieval, and graph infrastructure for cognitive systems
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

//! Linear weighted fusion.
//!
//! Combines normalised BM25 and vector scores by weighted sum:
//!
//! ```text
//! score(d) = bm25_weight * normalise(bm25_score(d))
//!           + vector_weight * normalise(vector_score(d))
//! ```
//!
//! Documents that appear in only one list are scored with 0.0 for the
//! missing component.

use std::collections::HashMap;

use crate::hybrid::{
    config::NormalisationMethod,
    fusion::{FusedScore, FusionStrategy, ScoredItem},
    normalizer,
};

/// Linear fusion strategy.
#[derive(Debug, Clone)]
pub struct Linear {
    /// BM25 list weight.
    pub bm25_weight: f32,
    /// Vector list weight.
    pub vector_weight: f32,
    /// Score normalisation applied before combining.
    pub normalisation: NormalisationMethod,
}

impl Linear {
    pub fn new(bm25_weight: f32, vector_weight: f32, normalisation: NormalisationMethod) -> Self {
        Self { bm25_weight, vector_weight, normalisation }
    }
}

impl Default for Linear {
    fn default() -> Self {
        Self::new(0.5, 0.5, NormalisationMethod::MinMax)
    }
}

impl FusionStrategy for Linear {
    fn fuse(
        &self,
        bm25_items: &[ScoredItem],
        vector_items: &[ScoredItem],
    ) -> HashMap<String, FusedScore> {
        // Convert to (id, score) pairs for normalisation
        let bm25_pairs: Vec<(String, f32)> = bm25_items
            .iter()
            .map(|item| (item.id.clone(), item.score))
            .collect();
        let vector_pairs: Vec<(String, f32)> = vector_items
            .iter()
            .map(|item| (item.id.clone(), item.score))
            .collect();

        let norm_bm25 = self.normalise(&bm25_pairs);
        let norm_vector = self.normalise(&vector_pairs);

        // Union of all ids
        let all_ids: std::collections::HashSet<String> = bm25_pairs
            .iter()
            .map(|(id, _)| id.clone())
            .chain(vector_pairs.iter().map(|(id, _)| id.clone()))
            .collect();

        all_ids
            .into_iter()
            .map(|id| {
                let b = norm_bm25.get(&id).copied().unwrap_or(0.0);
                let v = norm_vector.get(&id).copied().unwrap_or(0.0);
                let hybrid = self.bm25_weight * b + self.vector_weight * v;
                let fused = FusedScore {
                    hybrid,
                    bm25: bm25_pairs.iter().find(|(bid, _)| bid == &id).map(|(_, s)| *s),
                    vector: vector_pairs.iter().find(|(vid, _)| vid == &id).map(|(_, s)| *s),
                };
                (id, fused)
            })
            .collect()
    }
}

impl Linear {
    fn normalise(&self, pairs: &[(String, f32)]) -> normalizer::NormalisedScores {
        match self.normalisation {
            NormalisationMethod::MinMax => normalizer::min_max(pairs),
            NormalisationMethod::ZScore => normalizer::z_score(pairs),
            NormalisationMethod::None => normalizer::none(pairs),
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    fn items(ids: &[&str], scores: &[f32]) -> Vec<ScoredItem> {
        ids.iter()
            .zip(scores.iter())
            .map(|(id, s)| ScoredItem::new(*id, *s))
            .collect()
    }

    // --- basic correctness ---

    #[test]
    fn empty_inputs_produce_empty_output() {
        let linear = Linear::default();
        assert!(linear.fuse(&[], &[]).is_empty());
    }

    #[test]
    fn bm25_only_with_no_weight_is_zero() {
        let linear = Linear::new(0.0, 1.0, NormalisationMethod::None);
        let bm25 = items(&["a"], &[1.0]);
        let result = linear.fuse(&bm25, &[]);
        assert!((result["a"].hybrid).abs() < 1e-6);
    }

    #[test]
    fn vector_only_with_no_weight_is_zero() {
        let linear = Linear::new(1.0, 0.0, NormalisationMethod::None);
        let vec = items(&["a"], &[1.0]);
        let result = linear.fuse(&[], &vec);
        assert!((result["a"].hybrid).abs() < 1e-6);
    }

    #[test]
    fn document_in_both_lists_combines_scores() {
        let linear = Linear::new(0.5, 0.5, NormalisationMethod::None);
        let bm25 = items(&["a"], &[0.8]);
        let vec = items(&["a"], &[0.6]);
        let result = linear.fuse(&bm25, &vec);
        // 0.5 * 0.8 + 0.5 * 0.6 = 0.7
        assert!((result["a"].hybrid - 0.7).abs() < 1e-5);
    }

    #[test]
    fn document_only_in_bm25_gets_zero_vector_contribution() {
        let linear = Linear::new(0.5, 0.5, NormalisationMethod::None);
        let bm25 = items(&["a"], &[1.0]);
        let vec = items(&["b"], &[1.0]);
        let result = linear.fuse(&bm25, &vec);
        // "a" has vector = 0: score = 0.5 * 1.0 + 0.5 * 0.0 = 0.5
        assert!((result["a"].hybrid - 0.5).abs() < 1e-5);
    }

    // --- normalisation ---

    #[test]
    fn min_max_normalisation_maps_to_unit_range() {
        let linear = Linear::new(1.0, 0.0, NormalisationMethod::MinMax);
        let bm25 = items(&["lo", "hi"], &[0.0, 10.0]);
        let result = linear.fuse(&bm25, &[]);
        // After min-max: lo → 0, hi → 1 × weight 1.0
        assert!((result["lo"].hybrid).abs() < 1e-5);
        assert!((result["hi"].hybrid - 1.0).abs() < 1e-5);
    }

    #[test]
    fn z_score_normalisation_centres_at_zero() {
        let linear = Linear::new(1.0, 0.0, NormalisationMethod::ZScore);
        let bm25 = items(&["a", "b", "c"], &[1.0, 2.0, 3.0]);
        let result = linear.fuse(&bm25, &[]);
        // Mean = 2.0; b (score=2.0) should have hybrid ≈ 0.0 × weight
        assert!(result["b"].hybrid.abs() < 1e-4);
    }

    #[test]
    fn no_normalisation_uses_raw_scores() {
        let linear = Linear::new(1.0, 0.0, NormalisationMethod::None);
        let bm25 = items(&["a"], &[3.5]);
        let result = linear.fuse(&bm25, &[]);
        assert!((result["a"].hybrid - 3.5).abs() < 1e-5);
    }

    // --- score passthrough ---

    #[test]
    fn raw_bm25_score_preserved_in_fused_score() {
        let linear = Linear::new(0.5, 0.5, NormalisationMethod::None);
        let bm25 = items(&["a"], &[7.0]);
        let result = linear.fuse(&bm25, &[]);
        assert_eq!(result["a"].bm25, Some(7.0));
        assert!(result["a"].vector.is_none());
    }

    #[test]
    fn raw_vector_score_preserved_in_fused_score() {
        let linear = Linear::new(0.5, 0.5, NormalisationMethod::None);
        let vec = items(&["a"], &[0.9]);
        let result = linear.fuse(&[], &vec);
        assert_eq!(result["a"].vector, Some(0.9));
        assert!(result["a"].bm25.is_none());
    }

    // --- union behaviour ---

    #[test]
    fn union_of_ids_is_present_in_output() {
        let linear = Linear::default();
        let bm25 = items(&["a", "b"], &[1.0, 0.5]);
        let vec = items(&["b", "c"], &[0.9, 0.3]);
        let result = linear.fuse(&bm25, &vec);
        assert!(result.contains_key("a"));
        assert!(result.contains_key("b"));
        assert!(result.contains_key("c"));
    }
}