vicinity 0.11.0

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
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

from __future__ import annotations

import importlib.util
import json
import struct
import sys
from pathlib import Path
from types import ModuleType


def load_script(name: str) -> ModuleType:
    script_path = Path(__file__).resolve().parents[1] / f"scripts/{name}.py"
    spec = importlib.util.spec_from_file_location(name, script_path)
    assert spec is not None
    assert spec.loader is not None
    module = importlib.util.module_from_spec(spec)
    sys.modules[spec.name] = module
    spec.loader.exec_module(module)
    return module


def write_smoke_dataset(path: Path) -> None:
    smoke = load_script("generate_ann_smoke_data")
    path.mkdir()
    train = smoke.make_train(32, 6)
    test = smoke.make_test(train, 8)
    neighbors = smoke.ground_truth(train, test, 10)
    smoke.write_vec1(path / "train.bin", train)
    smoke.write_vec1(path / "test.bin", test)
    smoke.write_nbr1(path / "neighbors.bin", neighbors)


def write_lbl1(path: Path, labels: list[int]) -> None:
    with path.open("wb") as f:
        f.write(b"LBL1")
        f.write(struct.pack("<I", len(labels)))
        f.write(struct.pack(f"<{len(labels)}I", *labels))


def test_profile_dataset_reports_shape_and_difficulty(tmp_path: Path) -> None:
    script = load_script("profile_ann_dataset")
    dataset = tmp_path / "smoke-angular"
    write_smoke_dataset(dataset)

    profile = script.profile_dataset(
        dataset,
        metric="cosine",
        sample_train=16,
        sample_queries=4,
        pair_samples=20,
        coarse_clusters=4,
        coarse_iters=3,
        lid_k=5,
        seed=1,
    )

    assert profile["shape"] == {
        "dim": 6,
        "ground_truth_k": 10,
        "queries": 8,
        "train": 32,
    }
    assert profile["metric"] == "cosine"
    assert profile["query_splits"] == [
        {
            "dim": 6,
            "ground_truth_k": 10,
            "kind": "in_distribution",
            "name": "base",
            "queries": 8,
        }
    ]
    assert profile["samples"]["train"] == 16
    assert profile["samples"]["queries"] == 4
    assert profile["norms"]["train"]["p50"] > 0.99
    dispersion = profile["coordinate_dispersion_sample"]
    assert dispersion["centroid_norm"] > 0.0
    assert dispersion["dimension_std"]["p50"] >= 0.0
    assert 0.0 < dispersion["variance_effective_dim"] <= 6.0
    assert 0.0 < dispersion["variance_effective_dim_fraction"] <= 1.0
    assert 0.0 <= dispersion["zero_variance_fraction"] <= 1.0
    assert profile["query_neighbors"]["nearest_distance"]["mean"] >= 0.0
    assert profile["query_neighbors"]["top2_gap"]["mean"] >= 0.0
    assert profile["query_neighbors"]["lid_mle"]["mean"] > 0.0
    assert profile["sampled_relative_contrast"]["mean"] > 1.0
    assert profile["hubness"]["topk"] == 10.0
    assert profile["coarse_partition_imbalance"]["clusters"] == 4.0
    assert profile["coarse_partition_imbalance"]["count_max"] > 0.0


def test_cli_writes_json_output(tmp_path: Path, monkeypatch) -> None:
    script = load_script("profile_ann_dataset")
    dataset = tmp_path / "smoke-angular"
    output = tmp_path / "profile.json"
    write_smoke_dataset(dataset)
    monkeypatch.setattr(
        sys,
        "argv",
        [
            "profile_ann_dataset.py",
            str(dataset),
            "--sample-train",
            "16",
            "--sample-queries",
            "4",
            "--pair-samples",
            "20",
            "--output",
            str(output),
        ],
    )

    script.main()

    profile = json.loads(output.read_text(encoding="utf-8"))
    assert profile["dataset"] == str(dataset)
    assert profile["metric"] == "cosine"


def test_profile_dataset_reports_optional_splits(tmp_path: Path) -> None:
    script = load_script("profile_ann_dataset")
    smoke = load_script("generate_ann_smoke_data")
    dataset = tmp_path / "smoke-angular"
    write_smoke_dataset(dataset)
    train = smoke.make_train(32, 6)
    test = smoke.make_test(train, 8)
    neighbors = smoke.ground_truth(train, test, 10)
    smoke.write_vec1(dataset / "test_drift.bin", test)
    smoke.write_nbr1(dataset / "neighbors_drift.bin", neighbors)
    smoke.write_vec1(dataset / "test_filter.bin", test)
    smoke.write_nbr1(dataset / "neighbors_filter.bin", neighbors)
    write_lbl1(dataset / "test_filter_topics.bin", [0, 1, 0, 1, 0, 1, 0, 1])
    write_lbl1(dataset / "test_difficulty.bin", [0, 0, 1, 1, 1, 2, 2, 2])

    profile = script.profile_dataset(
        dataset,
        metric="cosine",
        sample_train=16,
        sample_queries=4,
        pair_samples=20,
        coarse_clusters=4,
        coarse_iters=3,
        lid_k=5,
        seed=1,
    )

    assert profile["query_splits"] == [
        {
            "difficulty_labels": "test_difficulty.bin",
            "dim": 6,
            "ground_truth_k": 10,
            "kind": "in_distribution",
            "name": "base",
            "queries": 8,
        },
        {
            "dim": 6,
            "ground_truth_k": 10,
            "kind": "ood_drift",
            "name": "drift",
            "queries": 8,
        },
        {
            "dim": 6,
            "ground_truth_k": 10,
            "kind": "filtered",
            "label_file": "test_filter_topics.bin",
            "name": "filter",
            "queries": 8,
        },
    ]


def test_infer_metric_uses_manifest(tmp_path: Path) -> None:
    script = load_script("profile_ann_dataset")
    dataset = tmp_path / "custom"
    dataset.mkdir()
    (dataset / "dataset.json").write_text(
        json.dumps({"settings": {"metric": "euclidean"}}),
        encoding="utf-8",
    )

    assert script.infer_metric(dataset, "auto") == "l2"


def test_truncated_vec1_is_rejected(tmp_path: Path) -> None:
    script = load_script("profile_ann_dataset")
    path = tmp_path / "train.bin"
    path.write_bytes(b"VEC1" + (2).to_bytes(4, "little") + (3).to_bytes(4, "little"))

    try:
        script.read_vec1(path)
    except SystemExit as exc:
        assert "expected 36" in str(exc)
    else:
        raise AssertionError("truncated VEC1 should be rejected")