def test_concepts_performance(events_service, concepts_service, test_results):
input_dir = Path(os.environ['BIOMEDICUS_TEST_DATA']) / 'concepts'
recall = Accuracy(name='recall', mode='any', fields=['cui'])
precision = Accuracy(name='precision', mode='any', fields=['cui'])
with EventsClient(address=events_service) as client, \
Pipeline(
RemoteProcessor(processor_id='biomedicus-concepts', address=concepts_service),
LocalProcessor(Metrics(recall, tested='umls_concepts', target='gold_concepts'),
component_id='metrics'),
LocalProcessor(Metrics(precision, tested='gold_concepts', target='umls_concepts'),
component_id='metrics_reverse'),
events_client=client
) as pipeline:
for test_file in input_dir.glob('**/*.pickle'):
with PickleSerializer.file_to_event(test_file, client=client) as event:
document = event.documents['plaintext']
pipeline.run(document)
print('Precision:', precision.value)
print('Recall:', recall.value)
timing_info = pipeline.processor_timer_stats('biomedicus-concepts').timing_info
test_results['Concepts'] = {
'Precision': precision.value,
'Recall': recall.value,
'Remote Call Duration': str(timing_info['remote_call'].mean),
'Process Method Duration': str(timing_info['process_method'].mean)
}
assert recall.value > 0.6
def test_acronyms_performance(events_service, acronyms_service, test_results):
input_dir = Path(os.environ['BIOMEDICUS_PHI_TEST_DATA']) / 'acronyms'
top_score_accuracy = Accuracy(name='top_score_accuracy',
fields=['expansion'])
any_accuracy = Accuracy(name='any_accuracy',
mode='any',
fields=['expansion'])
detection_recall = Accuracy(name='detection_recall',
mode='location',
fields=['expansion'])
detection_precision = Accuracy(name='detection_precision',
mode='location',
fields=['expansion'])
with EventsClient(address=events_service) as client, Pipeline(
RemoteProcessor(processor_id='biomedicus-acronyms',
address=acronyms_service),
LocalProcessor(Metrics(top_score_accuracy,
detection_recall,
tested='acronyms',
target='gold_acronyms'),
component_id='top_score_metrics',
client=client),
LocalProcessor(Metrics(detection_precision,
tested='gold_acronyms',
target='acronyms'),
component_id='top_score_reverse',
client=client),
LocalProcessor(Metrics(any_accuracy,
tested='all_acronym_senses',
target='gold_acronyms'),
component_id='all_senses_metrics',
client=client)) as pipeline:
for test_file in input_dir.glob('**/*.json'):
with JsonSerializer.file_to_event(test_file,
client=client) as event:
document = event.documents['plaintext']
pipeline.run(document)
print('Top Sense Accuracy:', top_score_accuracy.value)
print('Any Sense Accuracy:', any_accuracy.value)
print('Detection Recall:', detection_recall.value)
print('Detection Precision:', detection_precision.value)
pipeline.print_times()
timing_info = pipeline.processor_timer_stats(
'biomedicus-acronyms').timing_info
test_results['acronyms'] = {
'Top sense accuracy': top_score_accuracy.value,
'Any sense accuracy': any_accuracy.value,
'Detection Recall': detection_recall.value,
'Detection Precision': detection_precision.value,
'Remote Call Duration': str(timing_info['remote_call'].mean),
'Process Method Duration': str(timing_info['process_method'].mean)
}
assert top_score_accuracy.value > 0.4
assert any_accuracy.value > 0.4
assert detection_recall.value > 0.65
def test_fields():
with Event(event_id='1') as event:
doc = event.create_document('test', 'This is some text.')
with doc.get_labeler('tested') as tested:
tested(0, 5, x=1, y=3)
tested(6, 10, x=3, y=4)
with doc.get_labeler('target') as target:
target(0, 5, x=1, y=5)
target(6, 10, x=2, y=6)
acc = Accuracy(fields=['x'])
metrics = Metrics(acc, tested='tested', target='target')
metrics.process_document(doc, params={})
assert abs(acc.value - 0.5) < 1e-6
def test_any():
with Event(event_id='1') as event:
doc = event.create_document('test', 'This is some text.')
with doc.get_labeler('tested') as tested:
tested(0, 5, x=1)
tested(0, 5, x=3)
with doc.get_labeler('target') as target:
target(0, 5, x=1)
target(6, 10, x=2)
acc = Accuracy(mode='any')
metrics = Metrics(acc, tested='tested', target='target')
metrics.process_document(doc, params={})
assert abs(acc.value - 0.5) < 1e-6
def test_tnt_performance(events_service, pos_tags_service, test_results):
input_dir = Path(os.environ['BIOMEDICUS_TEST_DATA']) / 'pos_tags'
accuracy = Accuracy()
with EventsClient(address=events_service) as client, Pipeline(
RemoteProcessor(processor_id='biomedicus-tnt-tagger',
address=pos_tags_service,
params={'token_index': 'gold_tags'}),
LocalProcessor(Metrics(accuracy,
tested='pos_tags',
target='gold_tags'),
component_id='metrics'),
events_client=client) as pipeline:
for test_file in input_dir.glob('**/*.pickle'):
event = PickleSerializer.file_to_event(test_file, client=client)
with event:
document = event.documents['gold']
results = pipeline.run(document)
print(
'Accuracy for event - ', event.event_id, ':',
results.component_result(
'metrics').result_dict['accuracy'])
print('Accuracy:', accuracy.value)
pipeline.print_times()
timing_info = pipeline.processor_timer_stats(
'biomedicus-tnt-tagger').timing_info
test_results['TnT Pos Tagger'] = {
'Accuracy': accuracy.value,
'Remote Call Duration': str(timing_info['remote_call'].mean),
'Process Method Duration': str(timing_info['process_method'].mean)
}
assert accuracy.value > 0.9
def test_dependencies(events_service, dependencies_service, test_results):
test_dir = Path(os.environ['BIOMEDICUS_TEST_DATA']) / 'dependencies'
uas = Accuracy('UAS', equivalence_test=uas_equal)
las = Accuracy('LAS', equivalence_test=las_equal)
with EventsClient(address=events_service) as client, \
Pipeline(
RemoteProcessor(processor_id='biomedicus-dependencies',
address=dependencies_service),
LocalProcessor(Metrics(uas, las, tested='dependencies', target='gold_dependencies'),
component_id='accuracy', client=client)
) as pipeline:
for test_file in test_dir.glob('**/*.pickle'):
with PickleSerializer.file_to_event(test_file,
client=client) as event:
document = event.documents['plaintext']
results = pipeline.run(document)
accuracy_dict = results.component_result(
'accuracy').result_dict
print('Results for document: UAS: {}. LAS: {}.'.format(
accuracy_dict['UAS'], accuracy_dict['LAS']))
print('UAS:', uas.value)
print('LAS:', las.value)
timing_info = pipeline.processor_timer_stats(
'biomedicus-dependencies').timing_info
test_results['biomedicus-dependencies'] = {
'UAS': uas.value,
'LAS': las.value,
'Corpus': "MiPACQ converted to UD from PTB test set",
'Remote Call Duration': str(timing_info['remote_call'].mean),
'Process Method Duration': str(timing_info['process_method'].mean)
}
def test_boundary_fuzz_equals():
with Event(event_id='1') as event:
doc = event.create_document('test', 'This is some text.')
with doc.get_labeler('tested') as tested:
tested(0, 5)
tested(6, 10)
tested(11, 20)
tested(21, 29)
tested(31, 39)
with doc.get_labeler('target') as target:
target(0, 6)
target(7, 10)
target(11, 19)
target(20, 30)
target(49, 50)
acc = Accuracy(boundary_fuzz=1)
metrics = Metrics(acc, tested='tested', target='target')
metrics.process_document(doc, params={})
assert abs(acc.value - 0.8) < 1e-6
