💡 All-in-one AI framework for semantic search, LLM orchestration and language model workflows
Python agentsaiai-agentsembeddingsinformation-retrievallanguage-modellarge-language-modelsllmnlppythonragretrieval-augmented-generationsearchsearch-enginesemantic-searchsentence-embeddingstransformerstxtaivector-databasevector-search
"""
Labels module tests
"""
import unittest
from txtai.pipeline import Labels
class TestLabels(unittest.TestCase):
"""
Labels tests.
"""
@classmethod
def setUpClass(cls):
"""
Create single labels instance.
"""
cls.data = [
"US tops 5 million confirmed virus cases",
"Canada's last fully intact ice shelf has suddenly collapsed, forming a Manhattan-sized iceberg",
"Beijing mobilises invasion craft along coast as Taiwan tensions escalate",
"The National Park Service warns against sacrificing slower friends in a bear attack",
"Maine man wins $1M from $25 lottery ticket",
"Make huge profits without work, earn up to $100,000 a day",
]
cls.labels = Labels("prajjwal1/bert-medium-mnli")
def testLabel(self):
"""
Test labels with single text input
"""
self.assertEqual(self.labels("This is the best sentence ever", ["positive", "negative"])[0][0], 0)
def testLabelFlatten(self):
"""
Test labels with single text input, flattened to top text labels
"""
self.assertEqual(self.labels("This is the best sentence ever", ["positive", "negative"], flatten=True)[0], "positive")
def testLabelBatch(self):
"""
Test labels with multiple text inputs
"""
results = [l[0][0] for l in self.labels(["This is the best sentence ever", "This is terrible"], ["positive", "negative"])]
self.assertEqual(results, [0, 1])
def testLabelBatchFlatten(self):
"""
Test labels with multiple text inputs, flattened to top text labels
"""
results = [l[0] for l in self.labels(["This is the best sentence ever", "This is terrible"], ["positive", "negative"], flatten=True)]
self.assertEqual(results, ["positive", "negative"])
def testLabelZeroThreshold(self):
"""
Test numeric zero flattening thresholds with dynamic and fixed labels
"""
text = "This is the best sentence ever"
for classifier, candidates in ((self.labels, ["positive", "negative"]), (Labels(dynamic=False), ["POSITIVE", "NEGATIVE"])):
with self.subTest(dynamic=classifier.dynamic):
self.assertCountEqual(classifier(text, candidates, flatten=0.0), candidates)
self.assertCountEqual(classifier([text], candidates, flatten=0.0)[0], candidates)
# Boolean False still requests the original score tuples.
self.assertEqual(classifier(text, candidates, flatten=False), classifier(text, candidates))
def testLabelFixed(self):
"""
Test labels with a fixed label text classification model
"""
labels = Labels(dynamic=False)
# Get index of "POSITIVE" label
index = labels.labels().index("POSITIVE")
# Verify results
self.assertEqual(labels("This is the best sentence ever")[0][0], index)
self.assertEqual(labels("This is the best sentence ever", multilabel=True)[0][0], index)
def testLabelFixedFlatten(self):
"""
Test labels with a fixed label text classification model, flattened to top text labels
"""
labels = Labels(dynamic=False)
# Verify results
self.assertEqual(labels("This is the best sentence ever", flatten=True)[0], "POSITIVE")
self.assertEqual(labels("This is the best sentence ever", multilabel=True, flatten=True)[0], "POSITIVE")
def testLabelFixedLimit(self):
"""
Test labels with a fixed label text classification model, limited to requested labels by name and by id
"""
labels = Labels(dynamic=False)
# Rotate the config label mappings so insertion order no longer matches label ids
config = labels.pipeline.model.config
ids = list(config.id2label)
config.id2label = {x: config.id2label[x] for x in ids[1:] + ids[:1]}
config.label2id = {label: x for x, label in config.id2label.items()}
# Verify requested labels resolve to label ids, not positions in the config label mappings
for label, uid in config.label2id.items():
self.assertEqual(labels("This is the best sentence ever", labels=[label.lower()])[0][0], uid)
self.assertEqual(labels("This is the best sentence ever", labels=[str(uid)])[0][0], uid)