def add(self, entry):
self.add_metatype(entry.get('?metatype'), entry.get('where'))
self.add_type(entry.get('?type'))
self.add_mention(entry.get('?mention_span'),
trim_cv(entry.get('?type_statement_confidence')),
trim_cv(entry.get('?cluster_membership_confidence')),
trim_cv(entry.get('?mention_type_justification_confidence')),
entry.get('where'))
def normalize_confidences(responses, cluster_id):
max_confidence = None
for response in responses.values():
if response.get('cluster_id') != cluster_id: continue
justification_confidence = trim_cv(response.get('justification_confidence'))
if max_confidence is None:
max_confidence = justification_confidence
if justification_confidence > max_confidence:
max_confidence = justification_confidence
for response in responses.values():
normalized_confidence_value = trim_cv(response.get('justification_confidence'))/max_confidence
response.set('normalized_justification_confidence', normalized_confidence_value)
def compute_weights(responses, APPLY_NORMALIZATION, APPLY_WEIGHTS):
for response in responses.values():
weight = 1
if APPLY_WEIGHTS:
if APPLY_NORMALIZATION:
weight = response.get('normalized_justification_confidence')
else:
weight = trim_cv(response.get('justification_confidence'))
response.set('weight', weight)
def validate_importance_value(self, responses, schema, entry, attribute):
importance_value = entry.get(attribute.get('name'))
try:
value = trim_cv(importance_value)
except ValueError:
self.record_event('INVALID_IMPORTANCE_VALUE', importance_value,
entry.get('where'))
return False
return True
def get_aggregate_confidence(self, entry):
field_names = [
'type_statement_confidence', 'cluster_membership_confidence',
'mention_type_justification_confidence'
]
confidence = 1.0
for field_name in field_names:
confidence *= trim_cv(entry.get(field_name))
return confidence
def validate_confidence(self, responses, schema, entry, attribute):
value = trim_cv(entry.get(attribute.get('name')))
if schema.get('task') == 'task3' and schema.get('query_type') == 'GraphQuery' and value == 'NULL' and attribute.get('name') == 'edge_compound_justification_confidence':
return True
try:
float(value)
except ValueError:
entry.set(attribute.get('name'), 1)
self.record_event('INVALID_CONFIDENCE', value, entry.get('where'))
return False
if not 0 < float(value) <= 1:
self.record_event('INVALID_CONFIDENCE', value, entry.get('where'))
return False
return True
def validate_confidence(self, responses, schema, entry, attribute):
confidence_value = entry.get(attribute.get('name'))
if schema.get('task') == 'task3' and schema.get(
'name'
) == 'AIDA_PHASE2_TASK3_GR_RESPONSE' and attribute.get(
'name'
) == 'predicate_justification_confidence' and confidence_value == 'NULL':
return True
try:
value = trim_cv(confidence_value)
except ValueError:
self.record_event('INVALID_CONFIDENCE',
entry.get(attribute.get('name')),
entry.get('where'))
value = 1.0
entry.set(attribute.get('name'), '"{value}"'.format(value=value))
if not 0 < value <= 1:
self.record_event('INVALID_CONFIDENCE', value, entry.get('where'))
value = 1.0
entry.set(attribute.get('name'), '"{value}"'.format(value=value))
return True
