You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

267 lines
11 KiB

import pathlib
from unittest.mock import patch
from auto_labeling_pipeline.mappings import AmazonComprehendSentimentTemplate
from auto_labeling_pipeline.models import RequestModelFactory
from model_mommy import mommy
from rest_framework import status
from rest_framework.reverse import reverse
from api.models import DOCUMENT_CLASSIFICATION, IMAGE_CLASSIFICATION, SEQUENCE_LABELING
from api.models import Category, Span
from api.tests.api.utils import (CRUDMixin, make_auto_labeling_config, make_doc, make_image,
prepare_project)
data_dir = pathlib.Path(__file__).parent / 'data'
class TestTemplateList(CRUDMixin):
def setUp(self):
self.project = prepare_project(task=DOCUMENT_CLASSIFICATION)
self.url = reverse(viewname='auto_labeling_templates', args=[self.project.item.id])
def test_allow_admin_to_fetch_template_list(self):
self.url += '?task_name=DocumentClassification'
response = self.assert_fetch(self.project.users[0], status.HTTP_200_OK)
self.assertIn('Custom REST Request', response.data)
self.assertGreaterEqual(len(response.data), 1)
def test_deny_non_admin_to_fetch_template_list(self):
self.url += '?task_name=DocumentClassification'
for user in self.project.users[1:]:
self.assert_fetch(user, status.HTTP_403_FORBIDDEN)
def test_return_only_default_template_with_empty_task_name(self):
response = self.assert_fetch(self.project.users[0], status.HTTP_200_OK)
self.assertEqual(len(response.data), 1)
self.assertIn('Custom REST Request', response.data)
def test_return_only_default_template_with_wrong_task_name(self):
self.url += '?task_name=foobar'
response = self.assert_fetch(self.project.users[0], status.HTTP_200_OK)
self.assertEqual(len(response.data), 1)
self.assertIn('Custom REST Request', response.data)
class TestConfigParameter(CRUDMixin):
def setUp(self):
self.project = prepare_project(task=DOCUMENT_CLASSIFICATION)
self.data = {
'model_name': 'GCP Entity Analysis',
'model_attrs': {'key': 'hoge', 'type': 'PLAIN_TEXT', 'language': 'en'},
'text': 'example'
}
self.url = reverse(viewname='auto_labeling_parameter_testing', args=[self.project.item.id])
@patch('auto_labeling.views.RestAPIRequestTesting.send_request', return_value={})
def test_called_with_proper_model(self, mock):
self.assert_create(self.project.users[0], status.HTTP_200_OK)
_, kwargs = mock.call_args
expected = RequestModelFactory.create(self.data['model_name'], self.data['model_attrs'])
self.assertEqual(kwargs['model'], expected)
@patch('auto_labeling.views.RestAPIRequestTesting.send_request', return_value={})
def test_called_with_text(self, mock):
self.assert_create(self.project.users[0], status.HTTP_200_OK)
_, kwargs = mock.call_args
self.assertEqual(kwargs['example'], self.data['text'])
@patch('auto_labeling.views.RestAPIRequestTesting.send_request', return_value={})
def test_called_with_image(self, mock):
self.data['text'] = str(data_dir / 'images/1500x500.jpeg')
self.assert_create(self.project.users[0], status.HTTP_200_OK)
_, kwargs = mock.call_args
self.assertEqual(kwargs['example'], self.data['text'])
class TestTemplateMapping(CRUDMixin):
def setUp(self):
self.project = prepare_project(task=DOCUMENT_CLASSIFICATION)
self.data = {
'response': {
'Sentiment': 'NEUTRAL',
'SentimentScore': {
'Positive': 0.004438233096152544,
'Negative': 0.0005306027014739811,
'Neutral': 0.9950305223464966,
'Mixed': 5.80838445785048e-7
}
},
'template': AmazonComprehendSentimentTemplate().load()
}
self.url = reverse(viewname='auto_labeling_template_test', args=[self.project.item.id])
def test_template_mapping(self):
response = self.assert_create(self.project.users[0], status.HTTP_200_OK)
expected = [{'label': 'NEUTRAL'}]
self.assertEqual(response.json(), expected)
def test_json_decode_error(self):
self.data['template'] = ''
self.assert_create(self.project.users[0], status.HTTP_400_BAD_REQUEST)
class TestLabelMapping(CRUDMixin):
def setUp(self):
self.project = prepare_project(task=DOCUMENT_CLASSIFICATION)
self.data = {
'response': [{'label': 'NEGATIVE'}],
'label_mapping': {'NEGATIVE': 'Negative'}
}
self.url = reverse(viewname='auto_labeling_mapping_test', args=[self.project.item.id])
def test_label_mapping(self):
response = self.assert_create(self.project.users[0], status.HTTP_200_OK)
expected = [{'label': 'Negative'}]
self.assertEqual(response.json(), expected)
class TestConfigCreation(CRUDMixin):
def setUp(self):
self.project = prepare_project(task=DOCUMENT_CLASSIFICATION)
self.data = {
'model_name': 'Amazon Comprehend Sentiment Analysis',
'model_attrs': {
'aws_access_key': 'str',
'aws_secret_access_key': 'str',
'region_name': 'us-east-1',
'language_code': 'en'
},
'template': AmazonComprehendSentimentTemplate().load(),
'label_mapping': {'NEGATIVE': 'Negative'},
'task_type': 'Category'
}
self.url = reverse(viewname='auto_labeling_configs', args=[self.project.item.id])
def test_create_config(self):
response = self.assert_create(self.project.users[0], status.HTTP_201_CREATED)
self.assertEqual(response.data['model_name'], self.data['model_name'])
def test_list_config(self):
mommy.make('AutoLabelingConfig', project=self.project.item)
response = self.assert_fetch(self.project.users[0], status.HTTP_200_OK)
self.assertEqual(len(response.data), 1)
class TestAutoLabelingText(CRUDMixin):
def setUp(self):
self.project = prepare_project(task=DOCUMENT_CLASSIFICATION)
make_auto_labeling_config(self.project.item)
self.example = make_doc(self.project.item)
self.url = reverse(viewname='auto_labeling_annotation', args=[self.project.item.id, self.example.id])
@patch('auto_labeling.views.execute_pipeline', return_value=[])
def test_text_task(self, mock):
self.assert_create(self.project.users[0], status.HTTP_201_CREATED)
_, kwargs = mock.call_args
self.assertEqual(kwargs['text'], self.example.text)
class TestAutoLabelingImage(CRUDMixin):
def setUp(self):
self.project = prepare_project(task=IMAGE_CLASSIFICATION)
make_auto_labeling_config(self.project.item)
filepath = data_dir / 'images/1500x500.jpeg'
self.example = make_image(self.project.item, str(filepath))
self.url = reverse(viewname='auto_labeling_annotation', args=[self.project.item.id, self.example.id])
@patch('auto_labeling.views.execute_pipeline', return_value=[])
def test_text_task(self, mock):
self.assert_create(self.project.users[0], status.HTTP_201_CREATED)
_, kwargs = mock.call_args
expected = str(self.example.filename)
self.assertEqual(kwargs['text'], expected)
class TestAutomatedCategoryLabeling(CRUDMixin):
def setUp(self):
self.project = prepare_project(task=DOCUMENT_CLASSIFICATION, single_class_classification=False)
self.example = make_doc(self.project.item)
self.category_pos = mommy.make(
'CategoryType', project=self.project.item, text='POS'
)
self.category_neg = mommy.make(
'CategoryType', project=self.project.item, text='NEG'
)
self.url = reverse(viewname='automated_category_labeling', args=[self.project.item.id, self.example.id])
@patch('auto_labeling.views.execute_pipeline', return_value=[{'label': 'POS'}])
def test_category_labeling(self, mock):
mommy.make('AutoLabelingConfig', task_type='Category')
self.assert_create(self.project.users[0], status.HTTP_201_CREATED)
self.assertEqual(Category.objects.count(), 1)
self.assertEqual(Category.objects.first().label, self.category_pos)
@patch('auto_labeling.views.execute_pipeline', side_effect=[[{'label': 'POS'}], [{'label': 'NEG'}]])
def test_multiple_configs(self, mock):
mommy.make('AutoLabelingConfig', task_type='Category')
mommy.make('AutoLabelingConfig', task_type='Category')
self.assert_create(self.project.users[0], status.HTTP_201_CREATED)
self.assertEqual(Category.objects.count(), 2)
self.assertEqual(Category.objects.first().label, self.category_pos)
self.assertEqual(Category.objects.last().label, self.category_neg)
@patch('auto_labeling.views.execute_pipeline', side_effect=[[{'label': 'POS'}], [{'label': 'POS'}]])
def test_cannot_label_same_category_type(self, mock):
mommy.make('AutoLabelingConfig', task_type='Category')
mommy.make('AutoLabelingConfig', task_type='Category')
self.assert_create(self.project.users[0], status.HTTP_201_CREATED)
self.assertEqual(Category.objects.count(), 1)
class TestAutomatedSpanLabeling(CRUDMixin):
def setUp(self):
self.project = prepare_project(task=SEQUENCE_LABELING)
self.example = make_doc(self.project.item)
self.loc = mommy.make('SpanType', project=self.project.item, text='LOC')
self.url = reverse(viewname='automated_span_labeling', args=[self.project.item.id, self.example.id])
@patch('auto_labeling.views.execute_pipeline', return_value=[{'label': 'LOC', 'start_offset': 0, 'end_offset': 5}])
def test_span_labeling(self, mock):
mommy.make('AutoLabelingConfig', task_type='Span')
self.assert_create(self.project.users[0], status.HTTP_201_CREATED)
self.assertEqual(Span.objects.count(), 1)
self.assertEqual(Span.objects.first().label, self.loc)
@patch(
'auto_labeling.views.execute_pipeline',
side_effect=[
[{'label': 'LOC', 'start_offset': 0, 'end_offset': 5}],
[{'label': 'LOC', 'start_offset': 5, 'end_offset': 10}]
]
)
def test_multiple_configs(self, mock):
mommy.make('AutoLabelingConfig', task_type='Span')
mommy.make('AutoLabelingConfig', task_type='Span')
self.assert_create(self.project.users[0], status.HTTP_201_CREATED)
expected_spans = [
{'label': 'LOC', 'start_offset': 0, 'end_offset': 5},
{'label': 'LOC', 'start_offset': 5, 'end_offset': 10}
]
self.assertEqual(Span.objects.count(), len(expected_spans))
for actual, expected in zip(Span.objects.all(), expected_spans):
self.assertEqual(actual.label, self.loc)
self.assertEqual(actual.start_offset, expected['start_offset'])
self.assertEqual(actual.end_offset, expected['end_offset'])
@patch(
'auto_labeling.views.execute_pipeline',
side_effect=[
[{'label': 'LOC', 'start_offset': 0, 'end_offset': 5}],
[{'label': 'LOC', 'start_offset': 4, 'end_offset': 10}]
]
)
def test_cannot_label_overlapping_span(self, mock):
mommy.make('AutoLabelingConfig', task_type='Span')
mommy.make('AutoLabelingConfig', task_type='Span')
self.assert_create(self.project.users[0], status.HTTP_201_CREATED)
self.assertEqual(Span.objects.count(), 1)