vicinity 0.5.0

Approximate nearest-neighbor search
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

#![allow(clippy::unwrap_used, clippy::expect_used)]
//! Cross-crate integration tests for Tekne Stack.
//!
//! These tests verify that cross-crate dependencies work correctly:
//! - vicinity uses idpaq for ID compression
//! - All crates compile with their optional dependencies
//!
//! # Running These Tests
//!
//! ```bash
//! # Test idpaq integration
//! cargo test --features id-compression cross_crate
//! ```

// =============================================================================
// idpaq Integration
// =============================================================================

#[cfg(feature = "id-compression")]
mod idpaq_integration {
    use vicinity::compression::{DeltaVarintCompressor, IdSetCompressor};

    /// Verify that vicinity correctly delegates to idpaq for compression
    #[test]
    fn vicinity_uses_idpaq_compressor() {
        let compressor = DeltaVarintCompressor::new();

        // This pattern simulates HNSW neighbor lists
        let neighbor_ids: Vec<u32> = vec![42, 156, 789, 1234, 2345];
        let universe = 10_000;

        let compressed = compressor.compress_set(&neighbor_ids, universe).unwrap();
        let decompressed = compressor.decompress_set(&compressed, universe).unwrap();

        assert_eq!(neighbor_ids, decompressed, "Round-trip should preserve IDs");
    }

    /// Test compression with patterns typical of IVF posting lists
    #[test]
    fn ivf_posting_list_pattern() {
        let compressor = DeltaVarintCompressor::new();

        // IVF posting lists: many IDs from a large universe
        let posting_list: Vec<u32> = (0..500).map(|i| i * 200).collect();
        let universe = 100_000;

        let compressed = compressor.compress_set(&posting_list, universe).unwrap();
        let decompressed = compressor.decompress_set(&compressed, universe).unwrap();

        assert_eq!(posting_list, decompressed);

        // Verify meaningful compression
        let uncompressed_size = posting_list.len() * 4;
        let ratio = uncompressed_size as f64 / compressed.len() as f64;
        assert!(
            ratio > 1.5,
            "Should achieve some compression, got {:.2}x",
            ratio
        );
    }

    /// Test edge cases that might occur in ANN indexes
    #[test]
    fn ann_edge_cases() {
        let compressor = DeltaVarintCompressor::new();

        // Single neighbor (common in early HNSW layers)
        let single: Vec<u32> = vec![5000];
        let compressed = compressor.compress_set(&single, 10_000).unwrap();
        let decompressed = compressor.decompress_set(&compressed, 10_000).unwrap();
        assert_eq!(single, decompressed);

        // Maximum neighbors (M_max in HNSW)
        let max_neighbors: Vec<u32> = (0..64).map(|i| i * 100).collect();
        let compressed = compressor.compress_set(&max_neighbors, 10_000).unwrap();
        let decompressed = compressor.decompress_set(&compressed, 10_000).unwrap();
        assert_eq!(max_neighbors, decompressed);
    }
}

// =============================================================================
// Feature Compilation Tests
// =============================================================================

// =============================================================================
// Dependency Chain Verification
// =============================================================================

mod dependency_verification {
    /// Verify the SIMD dependency chain produces correct distance values.
    ///
    /// vicinity (innr feature, default) -> innr
    #[test]
    fn simd_distance_functions_produce_correct_values() {
        let a = [1.0_f32, 0.0, 0.0];
        let b = [0.707_f32, 0.707, 0.0];

        let cos_d = vicinity::distance::cosine_distance(&a, &b);
        let l2_d = vicinity::distance::l2_distance(&a, &b);

        // Expected cosine distance: 1 - dot(a,b)/(|a|*|b|) = 1 - 0.707/1.0 ≈ 0.293
        assert!(
            (cos_d - 0.293).abs() < 0.01,
            "cosine distance should be ~0.293, got {}",
            cos_d
        );
        // Expected L2: sqrt((1-0.707)^2 + 0.707^2) ≈ 0.586 (squared: ~0.586)
        assert!(
            l2_d > 0.1,
            "l2 distance should be meaningfully positive, got {}",
            l2_d
        );
    }
}