Spaces:
Runtime error
Runtime error
File size: 11,806 Bytes
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 |
"""Test for the concept scorer."""
import pathlib
from typing import Generator, Iterable, Type, cast
import numpy as np
import pytest
from pytest_mock import MockerFixture
from typing_extensions import override
from ..concepts.concept import ConceptColumnInfo, ConceptModel, ExampleIn
from ..concepts.db_concept import (
ConceptDB,
ConceptModelDB,
ConceptUpdate,
DiskConceptDB,
DiskConceptModelDB,
)
from ..config import CONFIG
from ..data.dataset_duckdb import DatasetDuckDB
from ..data.dataset_test_utils import TestDataMaker
from ..data.dataset_utils import lilac_embedding
from ..db_manager import set_default_dataset_cls
from ..embeddings.vector_store_numpy import NumpyVectorStore
from ..schema import UUID_COLUMN, Item, RichData, SignalInputType
from .concept_scorer import ConceptScoreSignal
from .signal import TextEmbeddingSignal, clear_signal_registry, register_signal
ALL_CONCEPT_DBS = [DiskConceptDB]
ALL_CONCEPT_MODEL_DBS = [DiskConceptModelDB]
@pytest.fixture(autouse=True)
def set_data_path(tmp_path: pathlib.Path, mocker: MockerFixture) -> None:
mocker.patch.dict(CONFIG, {'LILAC_DATA_PATH': str(tmp_path)})
EMBEDDING_MAP: dict[str, list[float]] = {
'not in concept': [0.1, 0.9, 0.0],
'in concept': [0.9, 0.1, 0.0],
'a new data point': [0.1, 0.2, 0.3],
'hello.': [0.1, 0.2, 0.3],
'hello2.': [0.1, 0.2, 0.3],
}
class TestEmbedding(TextEmbeddingSignal):
"""A test embed function."""
name = 'test_embedding'
@override
def compute(self, data: Iterable[RichData]) -> Iterable[Item]:
"""Embed the examples, use a hashmap to the vector for simplicity."""
for example in data:
if example not in EMBEDDING_MAP:
raise ValueError(f'Example "{str(example)}" not in embedding map')
yield [lilac_embedding(0, len(example), np.array(EMBEDDING_MAP[cast(str, example)]))]
@pytest.fixture(scope='module', autouse=True)
def setup_teardown() -> Generator:
# Setup.
set_default_dataset_cls(DatasetDuckDB)
register_signal(TestEmbedding)
# Unit test runs.
yield
# Teardown.
clear_signal_registry()
@pytest.mark.parametrize('db_cls', ALL_CONCEPT_DBS)
def test_embedding_does_not_exist(db_cls: Type[ConceptDB]) -> None:
db = db_cls()
namespace = 'test'
concept_name = 'test_concept'
db.create(namespace=namespace, name=concept_name, type=SignalInputType.TEXT)
train_data = [
ExampleIn(label=False, text='not in concept'),
ExampleIn(label=True, text='in concept')
]
db.edit(namespace, concept_name, ConceptUpdate(insert=train_data))
with pytest.raises(ValueError, match='Signal "unknown_embedding" not found in the registry'):
ConceptScoreSignal(namespace='test', concept_name='test_concept', embedding='unknown_embedding')
def test_concept_does_not_exist() -> None:
signal = ConceptScoreSignal(
namespace='test', concept_name='test_concept', embedding='test_embedding')
with pytest.raises(ValueError, match='Concept "test/test_concept" does not exist'):
signal.compute(['a new data point', 'not in concept'])
@pytest.mark.parametrize('concept_db_cls', ALL_CONCEPT_DBS)
@pytest.mark.parametrize('model_db_cls', ALL_CONCEPT_MODEL_DBS)
def test_concept_model_score(concept_db_cls: Type[ConceptDB],
model_db_cls: Type[ConceptModelDB]) -> None:
concept_db = concept_db_cls()
model_db = model_db_cls(concept_db)
namespace = 'test'
concept_name = 'test_concept'
concept_db.create(namespace=namespace, name=concept_name, type=SignalInputType.TEXT)
train_data = [
ExampleIn(label=False, text='not in concept'),
ExampleIn(label=True, text='in concept')
]
concept_db.edit(namespace, concept_name, ConceptUpdate(insert=train_data))
signal = ConceptScoreSignal(
namespace='test', concept_name='test_concept', embedding='test_embedding')
# Explicitly sync the model with the concept.
model = ConceptModel(
namespace='test', concept_name='test_concept', embedding_name='test_embedding')
model_db.sync(model, column_info=None)
scores = cast(list[float], list(signal.compute(['a new data point', 'not in concept'])))
assert scores[0] > 0 and scores[0] < 1
assert scores[1] < 0.5
@pytest.mark.parametrize('concept_db_cls', ALL_CONCEPT_DBS)
@pytest.mark.parametrize('model_db_cls', ALL_CONCEPT_MODEL_DBS)
def test_concept_model_with_dataset_score(concept_db_cls: Type[ConceptDB],
model_db_cls: Type[ConceptModelDB],
make_test_data: TestDataMaker) -> None:
dataset = make_test_data([{
UUID_COLUMN: '1',
'text': 'hello.',
}, {
UUID_COLUMN: '2',
'text': 'hello2.',
}])
concept_db = concept_db_cls()
model_db = model_db_cls(concept_db)
namespace = 'test'
concept_name = 'test_concept'
concept_db.create(namespace=namespace, name=concept_name, type=SignalInputType.TEXT)
train_data = [
ExampleIn(label=False, text='not in concept'),
ExampleIn(label=True, text='in concept')
]
concept_db.edit(namespace, concept_name, ConceptUpdate(insert=train_data))
column_info = ConceptColumnInfo(
namespace=dataset.namespace, name=dataset.dataset_name, path='text')
signal = ConceptScoreSignal(
namespace='test', concept_name='test_concept', embedding='test_embedding')
signal.set_column_info(column_info)
# Explicitly sync the model with the concept.
model = ConceptModel(
namespace='test', concept_name='test_concept', embedding_name='test_embedding')
model_db.sync(model, column_info)
scores = cast(list[float],
list(signal.compute(['a new data point', 'in concept', 'not in concept'])))
assert scores[0] > 0 and scores[0] < 1 # 'a new data point' may or may not be in the concept.
assert scores[1] > 0.5 # 'in concept' is in the concept.
assert scores[2] < 0.5 # 'not in concept' is not in the concept.
@pytest.mark.parametrize('concept_db_cls', ALL_CONCEPT_DBS)
@pytest.mark.parametrize('model_db_cls', ALL_CONCEPT_MODEL_DBS)
def test_concept_model_vector_score(concept_db_cls: Type[ConceptDB],
model_db_cls: Type[ConceptModelDB]) -> None:
concept_db = concept_db_cls()
model_db = model_db_cls(concept_db)
namespace = 'test'
concept_name = 'test_concept'
concept_db.create(namespace=namespace, name=concept_name, type=SignalInputType.TEXT)
train_data = [
ExampleIn(label=False, text='not in concept'),
ExampleIn(label=True, text='in concept')
]
concept_db.edit(namespace, concept_name, ConceptUpdate(insert=train_data))
signal = ConceptScoreSignal(
namespace='test', concept_name='test_concept', embedding='test_embedding')
# Explicitly sync the model with the concept.
model = ConceptModel(
namespace='test', concept_name='test_concept', embedding_name='test_embedding')
model_db.sync(model, column_info=None)
vector_store = NumpyVectorStore()
embeddings = np.array([
EMBEDDING_MAP['in concept'], EMBEDDING_MAP['not in concept'], EMBEDDING_MAP['a new data point']
])
vector_store.add([('1',), ('2',), ('3',)], embeddings)
scores = cast(list[float], list(signal.vector_compute([('1',), ('2',), ('3',)], vector_store)))
assert scores[0] > 0.5 # '1' is in the concept.
assert scores[1] < 0.5 # '2' is not in the concept.
assert scores[2] > 0 and scores[2] < 1 # '3' may or may not be in the concept.
@pytest.mark.parametrize('concept_db_cls', ALL_CONCEPT_DBS)
@pytest.mark.parametrize('model_db_cls', ALL_CONCEPT_MODEL_DBS)
def test_concept_model_topk_score(concept_db_cls: Type[ConceptDB],
model_db_cls: Type[ConceptModelDB]) -> None:
concept_db = concept_db_cls()
model_db = model_db_cls(concept_db)
namespace = 'test'
concept_name = 'test_concept'
concept_db.create(namespace=namespace, name=concept_name, type=SignalInputType.TEXT)
train_data = [
ExampleIn(label=False, text='not in concept'),
ExampleIn(label=True, text='in concept')
]
concept_db.edit(namespace, concept_name, ConceptUpdate(insert=train_data))
signal = ConceptScoreSignal(
namespace='test', concept_name='test_concept', embedding='test_embedding')
# Explicitly sync the model with the concept.
model = ConceptModel(
namespace='test', concept_name='test_concept', embedding_name='test_embedding')
model_db.sync(model, column_info=None)
vector_store = NumpyVectorStore()
vector_store.add([('1',), ('2',), ('3',)],
np.array([[0.1, 0.2, 0.3], [0.1, 0.87, 0.0], [1.0, 0.0, 0.0]]))
# Compute topk without id restriction.
topk_result = signal.vector_compute_topk(3, vector_store)
expected_result = [('3',), ('1',), ('2',)]
for (id, _), expected_id in zip(topk_result, expected_result):
assert id == expected_id
# Compute top 1.
topk_result = signal.vector_compute_topk(1, vector_store)
expected_result = [('3',)]
for (id, _), expected_id in zip(topk_result, expected_result):
assert id == expected_id
# Compute topk with id restriction.
topk_result = signal.vector_compute_topk(3, vector_store, keys=[('1',), ('2',)])
expected_result = [('1',), ('2',)]
for (id, _), expected_id in zip(topk_result, expected_result):
assert id == expected_id
@pytest.mark.parametrize('concept_db_cls', ALL_CONCEPT_DBS)
@pytest.mark.parametrize('model_db_cls', ALL_CONCEPT_MODEL_DBS)
def test_concept_model_draft(concept_db_cls: Type[ConceptDB],
model_db_cls: Type[ConceptModelDB]) -> None:
concept_db = concept_db_cls()
model_db = model_db_cls(concept_db)
namespace = 'test'
concept_name = 'test_concept'
concept_db.create(namespace=namespace, name=concept_name, type=SignalInputType.TEXT)
train_data = [
ExampleIn(label=False, text='not in concept'),
ExampleIn(label=True, text='in concept'),
ExampleIn(label=False, text='a new data point', draft='test_draft'),
]
concept_db.edit(namespace, concept_name, ConceptUpdate(insert=train_data))
signal = ConceptScoreSignal(
namespace='test', concept_name='test_concept', embedding='test_embedding')
draft_signal = ConceptScoreSignal(
namespace='test', concept_name='test_concept', embedding='test_embedding', draft='test_draft')
# Explicitly sync the model with the concept.
model = ConceptModel(
namespace='test', concept_name='test_concept', embedding_name='test_embedding')
model_db.sync(model, column_info=None)
vector_store = NumpyVectorStore()
vector_store.add([('1',), ('2',), ('3',)],
np.array([[1.0, 0.0, 0.0], [0.9, 0.1, 0.0], [0.1, 0.9, 0.0]]))
scores = cast(list[float], list(signal.vector_compute([('1',), ('2',), ('3',)], vector_store)))
assert scores[0] > 0.5
assert scores[1] > 0.5
assert scores[2] < 0.5
# Make sure the draft signal works. It has different values than the original signal.
vector_store = NumpyVectorStore()
vector_store.add([('1',), ('2',), ('3',)],
np.array([[1.0, 0.0, 0.0], [0.9, 0.1, 0.0], [0.1, 0.2, 0.3]]))
draft_scores = draft_signal.vector_compute([('1',), ('2',), ('3',)], vector_store)
assert draft_scores != scores
def test_concept_score_key() -> None:
signal = ConceptScoreSignal(
namespace='test', concept_name='test_concept', embedding=TestEmbedding.name)
assert signal.key() == 'test/test_concept'
@pytest.mark.parametrize('concept_db_cls', ALL_CONCEPT_DBS)
def test_concept_score_compute_signal_key(concept_db_cls: Type[ConceptDB]) -> None:
concept_db = concept_db_cls()
namespace = 'test'
concept_name = 'test_concept'
concept_db.create(namespace=namespace, name=concept_name, type=SignalInputType.TEXT)
signal = ConceptScoreSignal(
namespace='test', concept_name='test_concept', embedding=TestEmbedding.name)
assert signal.key(is_computed_signal=True) == 'test/test_concept/v0'
|