Spaces:
Runtime error
Runtime error
File size: 12,029 Bytes
e4f9cbe 55dc3dd e4f9cbe |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 |
"""Tests for dataset.select_rows(searches=[...])."""
from typing import Iterable, cast
import numpy as np
import pytest
from pytest import approx
from pytest_mock import MockerFixture
from sklearn.preprocessing import normalize
from typing_extensions import override
from ..concepts.concept import ExampleIn, LogisticEmbeddingModel
from ..concepts.db_concept import ConceptUpdate, DiskConceptDB
from ..db_manager import set_default_dataset_cls
from ..schema import UUID_COLUMN, Item, RichData, SignalInputType
from ..signals.concept_scorer import ConceptScoreSignal
from ..signals.semantic_similarity import SemanticSimilaritySignal
from ..signals.signal import TextEmbeddingSignal, clear_signal_registry, register_signal
from ..signals.substring_search import SubstringSignal
from .dataset import ConceptQuery, KeywordQuery, ListOp, Search, SemanticQuery, SortOrder
from .dataset_duckdb import DatasetDuckDB
from .dataset_test_utils import TestDataMaker, enriched_embedding_span, enriched_item
from .dataset_utils import lilac_embedding, lilac_span
TEST_DATA: list[Item] = [{
UUID_COLUMN: '1',
'text': 'hello world',
'text2': 'again hello world',
}, {
UUID_COLUMN: '2',
'text': 'looking for world in text',
'text2': 'again looking for world in text',
}, {
UUID_COLUMN: '3',
'text': 'unrelated text',
'text2': 'again unrelated text'
}]
EMBEDDINGS: list[tuple[str, list[float]]] = [
('hello.', [1.0, 0.0, 0.0]),
('hello2.', [1.0, 1.0, 0.0]),
('hello world.', [1.0, 1.0, 1.0]),
('hello world2.', [2.0, 1.0, 1.0]),
('random negative 1', [0, 0, 0.3]),
('random negative 2', [0, 0, 0.4]),
('random negative 3', [0, 0.1, 0.5]),
('random negative 4', [0.1, 0, 0.4]),
]
STR_EMBEDDINGS: dict[str, list[float]] = {text: embedding for text, embedding in EMBEDDINGS}
@pytest.fixture(scope='module', autouse=True)
def setup_teardown() -> Iterable[None]:
# Setup.
set_default_dataset_cls(DatasetDuckDB)
register_signal(TestEmbedding)
# Unit test runs.
yield
# Teardown.
clear_signal_registry()
def test_search_keyword(make_test_data: TestDataMaker) -> None:
dataset = make_test_data(TEST_DATA)
query = 'world'
result = dataset.select_rows(
searches=[Search(path='text', query=KeywordQuery(type='keyword', search=query))],
combine_columns=True)
expected_signal_udf = SubstringSignal(query=query)
assert list(result) == [{
UUID_COLUMN: '1',
'text': enriched_item('hello world', {expected_signal_udf.key(): [lilac_span(6, 11)]}),
'text2': 'again hello world'
}, {
UUID_COLUMN: '2',
'text': enriched_item('looking for world in text',
{expected_signal_udf.key(): [lilac_span(12, 17)]}),
'text2': 'again looking for world in text',
}]
def test_search_keyword_special_chars(make_test_data: TestDataMaker) -> None:
dataset = make_test_data([{
UUID_COLUMN: '1',
'text': 'This is 100%',
}, {
UUID_COLUMN: '2',
'text': 'This has _underscore_',
}])
query = '100%'
result = dataset.select_rows(
searches=[Search(path='text', query=KeywordQuery(type='keyword', search=query))],
combine_columns=True)
expected_signal_udf = SubstringSignal(query=query)
assert list(result) == [{
UUID_COLUMN: '1',
'text': enriched_item('This is 100%', {expected_signal_udf.key(): [lilac_span(8, 12)]}),
}]
query = '_underscore_'
result = dataset.select_rows(
searches=[Search(path='text', query=KeywordQuery(type='keyword', search=query))],
combine_columns=True)
expected_signal_udf = SubstringSignal(query=query)
assert list(result) == [{
UUID_COLUMN: '2',
'text': enriched_item('This has _underscore_',
{expected_signal_udf.key(): [lilac_span(9, 21)]}),
}]
def test_search_keyword_multiple(make_test_data: TestDataMaker) -> None:
dataset = make_test_data(TEST_DATA)
query_world = 'world'
query_looking_world = 'looking for world'
expected_world_udf = SubstringSignal(query=query_world)
expected_again_looking_udf = SubstringSignal(query=query_looking_world)
result = dataset.select_rows(
searches=[
Search(path='text', query=KeywordQuery(type='keyword', search=query_world)),
Search(path='text2', query=KeywordQuery(type='keyword', search=query_looking_world)),
],
combine_columns=True)
assert list(result) == [{
UUID_COLUMN: '2',
'text': enriched_item('looking for world in text', {
expected_world_udf.key(): [lilac_span(12, 17)],
}),
'text2': enriched_item('again looking for world in text',
{expected_again_looking_udf.key(): [lilac_span(6, 23)]})
}]
def test_search_keyword_with_filters(make_test_data: TestDataMaker) -> None:
dataset = make_test_data(TEST_DATA)
query = 'world'
result = dataset.select_rows(
filters=[(UUID_COLUMN, ListOp.IN, ['1', '3'])],
searches=[Search(path='text', query=KeywordQuery(type='keyword', search=query))],
combine_columns=True)
expected_signal_udf = SubstringSignal(query=query)
assert list(result) == [
{
UUID_COLUMN: '1',
'text': enriched_item('hello world', {expected_signal_udf.key(): [lilac_span(6, 11)]}),
'text2': 'again hello world'
},
# The second row doesn't match the UUID filter.
]
class TestEmbedding(TextEmbeddingSignal):
"""A test embed function."""
name = 'test_embedding'
@override
def compute(self, data: Iterable[RichData]) -> Iterable[Item]:
"""Call the embedding function."""
for example in data:
embedding = np.array(STR_EMBEDDINGS[cast(str, example)])
embedding = normalize([embedding])[0]
yield [lilac_embedding(0, len(example), embedding)]
def test_semantic_search(make_test_data: TestDataMaker) -> None:
dataset = make_test_data([{
UUID_COLUMN: '1',
'text': 'hello world.',
}, {
UUID_COLUMN: '2',
'text': 'hello world2.',
}])
test_embedding = TestEmbedding()
dataset.compute_signal(test_embedding, ('text'))
query = 'hello2.'
result = dataset.select_rows(
searches=[
Search(
path='text', query=SemanticQuery(type='semantic', search=query, embedding='test_embedding'))
],
combine_columns=True)
expected_signal_udf = SemanticSimilaritySignal(query=query, embedding='test_embedding')
assert list(result) == [
# Results are sorted by score desc.
{
UUID_COLUMN: '2',
'text': enriched_item(
'hello world2.', {
test_embedding.key():
[enriched_embedding_span(0, 13, {expected_signal_udf.key(): approx(0.916, 1e-3)})]
})
},
{
UUID_COLUMN: '1',
'text': enriched_item(
'hello world.', {
test_embedding.key():
[enriched_embedding_span(0, 12, {expected_signal_udf.key(): approx(0.885, 1e-3)})]
})
},
]
def test_concept_search(make_test_data: TestDataMaker, mocker: MockerFixture) -> None:
concept_model_mock = mocker.spy(LogisticEmbeddingModel, 'fit')
dataset = make_test_data([{
UUID_COLUMN: '1',
'text': 'hello world.',
}, {
UUID_COLUMN: '2',
'text': 'hello world2.',
}, {
UUID_COLUMN: '3',
'text': 'random negative 1',
}, {
UUID_COLUMN: '4',
'text': 'random negative 2',
}, {
UUID_COLUMN: '5',
'text': 'random negative 3',
}, {
UUID_COLUMN: '6',
'text': 'random negative 4',
}])
test_embedding = TestEmbedding()
dataset.compute_signal(test_embedding, ('text'))
concept_db = DiskConceptDB()
concept_db.create(namespace='test_namespace', name='test_concept', type=SignalInputType.TEXT)
concept_db.edit(
'test_namespace', 'test_concept',
ConceptUpdate(insert=[
ExampleIn(label=False, text='hello world.'),
ExampleIn(label=True, text='hello world2.')
]))
result = dataset.select_rows(
searches=[
Search(
path='text',
query=ConceptQuery(
type='concept',
concept_namespace='test_namespace',
concept_name='test_concept',
embedding='test_embedding'))
],
filters=[(UUID_COLUMN, ListOp.IN, ['1', '2'])],
combine_columns=True)
expected_signal_udf = ConceptScoreSignal(
namespace='test_namespace', concept_name='test_concept', embedding='test_embedding')
assert list(result) == [
# Results are sorted by score desc.
{
UUID_COLUMN: '2',
'text': enriched_item(
'hello world2.', {
test_embedding.key():
[enriched_embedding_span(0, 13, {expected_signal_udf.key(): approx(0.75, abs=0.25)})],
'test_namespace/test_concept/labels': [lilac_span(0, 13, {'label': True})]
})
},
{
UUID_COLUMN: '1',
'text': enriched_item(
'hello world.', {
test_embedding.key():
[enriched_embedding_span(0, 12, {expected_signal_udf.key(): approx(0.25, abs=0.25)})],
'test_namespace/test_concept/labels': [lilac_span(0, 12, {'label': False})]
})
},
]
(_, embeddings, labels, _) = concept_model_mock.call_args_list[-1].args
assert embeddings.shape == (2, 3)
assert labels == [
# Explicit labels.
False,
True
]
def test_sort_override_search(make_test_data: TestDataMaker) -> None:
dataset = make_test_data([{
UUID_COLUMN: '1',
'text': 'hello world.',
'value': 10
}, {
UUID_COLUMN: '2',
'text': 'hello world2.',
'value': 20
}])
test_embedding = TestEmbedding()
dataset.compute_signal(test_embedding, ('text'))
query = 'hello2.'
search = Search(
path='text', query=SemanticQuery(type='semantic', search=query, embedding='test_embedding'))
expected_signal_udf = SemanticSimilaritySignal(query=query, embedding='test_embedding')
expected_item_1 = {
UUID_COLUMN: '1',
'text': enriched_item(
'hello world.', {
test_embedding.key():
[enriched_embedding_span(0, 12, {expected_signal_udf.key(): approx(0.885, 1e-3)})]
}),
'value': 10
}
expected_item_2 = {
UUID_COLUMN: '2',
'text': enriched_item(
'hello world2.', {
test_embedding.key():
[enriched_embedding_span(0, 13, {expected_signal_udf.key(): approx(0.916, 1e-3)})]
}),
'value': 20
}
sort_order = SortOrder.ASC
result = dataset.select_rows(
searches=[search], sort_by=[('value',)], sort_order=sort_order, combine_columns=True)
assert list(result) == [
# Results are sorted by score ascending.
expected_item_1,
expected_item_2
]
sort_order = SortOrder.DESC
result = dataset.select_rows(
searches=[search], sort_by=[('text',)], sort_order=sort_order, combine_columns=True)
assert list(result) == [
# Results are sorted by score descending.
expected_item_2,
expected_item_1
]
def test_search_keyword_and_semantic(make_test_data: TestDataMaker) -> None:
dataset = make_test_data([{
UUID_COLUMN: '1',
'text': 'hello world.',
}, {
UUID_COLUMN: '2',
'text': 'hello world2.',
}])
test_embedding = TestEmbedding()
dataset.compute_signal(test_embedding, ('text'))
query = 'hello2.'
keyword_query = 'rld2'
result = dataset.select_rows(
searches=[
Search(
path='text', query=SemanticQuery(type='semantic', search=query,
embedding='test_embedding')),
Search(path='text', query=KeywordQuery(type='keyword', search=keyword_query))
],
combine_columns=True)
expected_semantic_signal = SemanticSimilaritySignal(query=query, embedding='test_embedding')
expected_keyword_signal = SubstringSignal(query=keyword_query)
assert list(result) == [
# Results are sorted by score desc.
{
UUID_COLUMN: '2',
'text': enriched_item(
'hello world2.', {
test_embedding.key():
[enriched_embedding_span(0, 13, {expected_semantic_signal.key(): approx(0.916, 1e-3)})],
expected_keyword_signal.key(): [lilac_span(8, 12)],
})
},
# UUID '1' is not returned because it does not match the keyword query.
]
|