hnsw-python/cat_hnsw/hnsw_consistent_build.py at master · generall/hnsw-python · GitHub

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
from itertools import groupby
from operator import itemgetter
from typing import Dict, Any, List, Iterable

import numpy as np
from tqdm import tqdm

from cat_hnsw.hnsw_cat import HNSWCat


class HNSWConsistentBuild(HNSWCat):
    """
    Preserve in-category connectivity.
    """

    def __init__(self, distance_type, m=5, ef=200, m0=None, heuristic=True, vectorized=False):
        super().__init__(distance_type, m, ef, m0, heuristic, vectorized)
        self._category_enter_points = {}

    @classmethod
    def _merge_layers(cls, layer_to, layer_from):
        for node, edges in layer_from.items():
            if node not in layer_to:
                layer_to[node] = edges
            else:
                layer_to[node].update(edges)

    def _merge_graphs(self, graphs: list):
        for layer_idx, layer in enumerate(graphs):
            if layer_idx == len(self._graphs):
                self._graphs.append(layer)
            else:
                self._merge_layers(self._graphs[layer_idx], layer)

    def build_cat_subgraph(self, points: Iterable[int], m, m0, ef):
        graphs = []
        entry_point = None

        for idx in points:
            entry_point = self._add(
                idx,
                data=self.data,
                graphs=graphs,
                entry_point=entry_point,
                m=m,
                m0=m0,
                ef=ef
            )
        self._merge_graphs(graphs)

        return graphs, entry_point

    def add_batch(
            self,
            data: np.ndarray,
            categories: Dict[Any, Iterable[int]] = None,
            connected_subsets: Iterable[Iterable[int]] = None,
            cat_m=None,
            cat_m0=None,
            subset_m=None,
            subset_m0=None,
            ef=None
    ):
        self.data = data
        for i in tqdm(range(self.data.shape[0])):
            self._enter_point = self._add(
                i,
                data=self.data,
                graphs=self._graphs,
                entry_point=self._enter_point,
                m=self._m,
                m0=self._m0,
                ef=ef
            )

        if categories is not None:
            cat_m = cat_m or self._m
            cat_m0 = cat_m0 or self._m0

            for category, points in tqdm(categories.items()):
                num_layers = len(self._graphs)
                graphs, entry_point = self.build_cat_subgraph(points, m=cat_m, m0=cat_m0, ef=ef)

                self._category_enter_points[category] = (entry_point, len(graphs) - 1)
                if len(self._graphs) > num_layers:
                    self._enter_point = entry_point

        if connected_subsets is not None:
            subset_m = subset_m or self._m
            subset_m0 = subset_m0 or self._m0

            for subset in connected_subsets:
                num_layers = len(self._graphs)
                graphs, entry_point = self.build_cat_subgraph(subset, m=subset_m, m0=subset_m0, ef=ef)
                if len(self._graphs) > num_layers:
                    self._enter_point = entry_point