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

use std::collections::HashSet;
use std::hash::Hash;
use std::fmt::Debug;

use rand::Rng;

use crate::hyperindex::HyperIndex;

pub struct DistanceNode<K: Eq+Hash> {
    pub key: K,
    pub distance: f32
}

impl<K:Eq+Hash> Eq for DistanceNode<K>
{
}

impl<K:Eq+Hash> PartialEq for DistanceNode<K> {
    fn eq(&self, other: &Self) -> bool {
        self.key == other.key
    }
}

impl<K:Eq+Hash> Hash for DistanceNode<K> {
    fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
        self.key.hash(state);
    }
}

pub struct MultiIndex<K:Send> {
    indices: Vec<HyperIndex<K>>
}

impl<K:Clone+Eq+Hash+Debug+Send> MultiIndex<K> {
    pub fn new<R : Rng + Sized>(dimension: usize, index_count: u8, hyperplane_count: u8, mut rng: &mut R) -> MultiIndex<K> {
        MultiIndex {
            indices: (0..index_count).map(|_| HyperIndex::new(dimension, hyperplane_count, &mut rng)).collect()
        }
    }

    pub fn nearest<F:Fn(&Vec<f32>, &K) -> f32>(&self, point: &Vec<f32>, count: usize, get_dist: F) -> Vec<DistanceNode<K>>
    {
        // Collect one group of results from every hyperindex
        // Dedupe by collecting into an intermediate hashset
        let mut results: Vec<_> = self.indices.iter()
            .flat_map(|i| i.group(&i.key(&point)))
            .flat_map(|a| a)
            .map(|a| DistanceNode { distance: get_dist(point, a), key: a.clone() })
            .collect::<HashSet<_>>()
            .into_iter()
            .collect::<Vec<_>>();

        // Sort into distance order and truncate to length
        results.sort_by(|a, b| a.distance.partial_cmp(&b.distance).unwrap_or(std::cmp::Ordering::Equal));
        results.truncate(count);

        return results;
    }

    pub fn add(&mut self, key: K, vector: &Vec<f32>) {

        // Build a list of work that needs doing (tuple of index, key and vector)
        let mut work:Vec<_> = self.indices.iter_mut()
            .map(|idx| (idx, &key, vector))
            .collect();

        //Add items in parallel
        work.iter_mut().for_each(|work| {
            work.0.add(work.1.clone(), work.2);
        });
    }

    pub fn dimensions(&self) -> usize {
        self.indices[0].dimensions()
    }

    pub fn planes_len(&self) -> usize {
        self.indices[0].planes_len()
    }

    pub fn indices_len(&self) -> usize {
        self.indices.len()
    }
}

#[cfg(test)]
mod tests
{
    use rand::prelude::*;
    use std::collections::HashSet;

    extern crate time;
    use time::Instant;

    use crate::multiindex::MultiIndex;
    use crate::vector::{ random_unit_vector, modified_cosine_distance, euclidean_distance };

    #[test]
    fn new_creates_index() {
        let a = MultiIndex::<usize>::new(300, 15, 10, &mut thread_rng());

        assert_eq!(300, a.dimensions());
        assert_eq!(10, a.planes_len());
        assert_eq!(15, a.indices_len());
    }

    #[test]
    fn multiindex_compare() {
        let mut a = MultiIndex::new(300, 30, 5, &mut thread_rng());

        let mut vectors = Vec::new();

        let mut rng = thread_rng();
        for key in 0..10000usize {
            let v = random_unit_vector(300, &mut rng);
            a.add(key, &v);
            vectors.push((key, v));
        }

        let start_linear = Instant::now();

        //Get closest points from simple search through the entire set
        println!();
        println!("Linear results:");
        let query_point = vectors[0].clone();
        let mut nearest_linear: Vec<(f32, &(usize, Vec<f32>))> = vectors.iter().map(|item| (euclidean_distance(&item.1, &query_point.1), item)).collect();
        nearest_linear.sort_by(|a, b| a.0.partial_cmp(&b.0).unwrap());

        let end_linear = Instant::now();
        println!("{:?} seconds for linear", end_linear - start_linear);

        for i in 0..20 {
            println!("idx:{:?}\t\tdist:{:?}", (nearest_linear[i].1).0, nearest_linear[i].0);
        }

        let start_indexed = Instant::now();

        //Use the index
        println!();
        println!("Index results:");
        let near = a.nearest(&query_point.1, 100, |p, k| {
            euclidean_distance(p, &vectors[*k].1)
        });

        let end_indexed = Instant::now();
        println!("{:?} seconds for index", end_indexed - start_indexed);
        
        for i in 0.. near.len().min(20) {
            println!("idx:{:?}\t\tdist:{:?}", near[i].key, near[i].distance);
        }

        let linear_set: HashSet<_> = nearest_linear.iter().map(|a| (a.1).0).take(20).collect();
        let near_set: HashSet<_> = near.iter().map(|a| a.key).take(20).collect();
        let overlap: Vec<_> = linear_set.intersection(&near_set).collect();
        println!();
        println!("Overlap:{:?}/20", overlap.len())
    }
}