def generate_document_id(self, responses, entry):
document_id = None
if entry.get('kb_document_id'):
document_id = entry.get('kb_document_id')
elif entry.get('object_informative_justification_span_text'):
span_object = spanstring_to_object(entry.get('logger'), entry.get('object_informative_justification_span_text'))
document_id = span_object.get('document_id')
elif entry.get('predicate_justification_span_text'):
span_object = spanstring_to_object(entry.get('logger'), entry.get('predicate_justification_span_text'))
document_id = span_object.get('document_id')
elif entry.get('mention_span_text'):
span_object = spanstring_to_object(entry.get('logger'), entry.get('mention_span_text'))
document_id = span_object.get('document_id')
entry.set('document_id', document_id)
def contains(self, mention, types):
contains = False
for cluster_type in types:
document_element_id = mention.get('document_element_id')
keyframe_id = mention.get('keyframe_id')
key = '{docid}:{doce_or_kf_id}:{cluster_type}'.format(
docid=mention.get('document_id'),
doce_or_kf_id=keyframe_id
if keyframe_id else document_element_id,
cluster_type=cluster_type)
if key not in self.get('regions'): continue
for span_string in self.get('regions').get(key):
span_string = '{docid}:{doce_or_kf_id}:{span_string}'.format(
docid=mention.get('document_id'),
doce_or_kf_id=keyframe_id
if keyframe_id else document_element_id,
span_string=span_string)
region = spanstring_to_object(self.get('logger'), span_string)
region.set(
'modality',
self.get('document_mappings').get(
'modality', region.get('document_element_id')))
boundaries_key = 'keyframe' if region.get(
'keyframe_id') else region.get('modality')
document_element_or_keyframe_id = region.get(
'keyframe_id') if region.get(
'keyframe_id') else region.get('document_element_id')
region.set(
'boundary',
self.get('document_boundaries').get(boundaries_key).get(
document_element_or_keyframe_id))
if get_intersection_over_union(mention, region) >= 0.8:
contains = True
return contains
def is_predicate_justification_correct(self, system_predicate_justifications, gold_predicate_justifications):
document_mappings = self.get('gold_responses').get('document_mappings')
document_boundaries = self.get('gold_responses').get('document_boundaries')
justification_correctness = False
max_num_justifications = 2
for system_predicate_justification in sorted(system_predicate_justifications.values(), key=lambda pj: pj.get('predicate_justification_confidence'), reverse=True):
system_predicate_justification_span = system_predicate_justification.get('predicate_justification')
system_mention_object = augment_mention_object(spanstring_to_object(self.logger, system_predicate_justification_span), document_mappings, document_boundaries)
for gold_predicate_justification in gold_predicate_justifications.values():
gold_predicate_justification_span = gold_predicate_justification.get('predicate_justification')
gold_mention_object = augment_mention_object(spanstring_to_object(self.logger, gold_predicate_justification_span), document_mappings, document_boundaries)
if get_intersection_over_union(system_mention_object, gold_mention_object) > 0:
justification_correctness = True
max_num_justifications -= 1
if max_num_justifications == 0: break
return justification_correctness
def add_mention(self, span_string, t_cv, cm_cv, j_cv, where):
logger = self.get('logger')
mention = augment_mention_object(spanstring_to_object(logger, span_string, where), self.get('document_mappings'), self.get('document_boundaries'))
mention.set('ID', span_string)
mention.set('span_string', span_string)
mention.set('t_cv', t_cv)
mention.set('cm_cv', cm_cv)
mention.set('j_cv', j_cv)
self.get('mentions').add(key=mention.get('ID'), value=mention)
def add_mention(self, span_string, t_cv, cm_cv, j_cv, where):
logger = self.get('logger')
mention = spanstring_to_object(logger, span_string, where)
mention.set('ID', span_string)
mention.set('span_string', span_string)
mention.set('t_cv', t_cv)
mention.set('cm_cv', cm_cv)
mention.set('j_cv', j_cv)
mention.set('modality', self.get('document_mappings').get('modality', mention.get('document_element_id')))
boundaries_key = 'keyframe' if mention.get('keyframe_id') else mention.get('modality')
document_element_or_keyframe_id = mention.get('keyframe_id') if mention.get('keyframe_id') else mention.get('document_element_id')
mention.set('boundary', self.get('document_boundaries').get(boundaries_key).get(document_element_or_keyframe_id))
self.get('mentions').add(key=mention.get('ID'), value=mention)
def get_mention(entry, document_mappings, document_boundaries):
logger = entry.get('logger')
where = entry.get('where')
span_string = entry.get('mention_span_text')
mention = spanstring_to_object(logger, span_string, where)
mention.set('ID', span_string)
mention.set('span_string', span_string)
mention.set('t_cv', entry.get('type_statement_confidence'))
mention.set('cm_cv', entry.get('cluster_membership_confidence'))
mention.set('j_cv', entry.get('mention_type_justification_confidence'))
mention.set('modality', document_mappings.get('modality', mention.get('document_element_id')))
boundaries_key = 'keyframe' if mention.get('keyframe_id') else mention.get('modality')
document_element_or_keyframe_id = mention.get('keyframe_id') if mention.get('keyframe_id') else mention.get('document_element_id')
mention.set('boundary', document_boundaries.get(boundaries_key).get(document_element_or_keyframe_id))
return mention
def contains_strict(self, mention, types, metatype):
document_element_id = mention.get('document_element_id')
keyframe_id = mention.get('keyframe_id')
for cluster_type in types:
key = '{docid}:{doce_or_kf_id}:{cluster_type}'.format(docid=mention.get('document_id'),
doce_or_kf_id=keyframe_id if keyframe_id else document_element_id,
cluster_type=cluster_type)
if key not in self.get('regions'): continue
for span_string in self.get('regions').get(key):
fq_span_string = '{docid}:{doce_or_kf_id}:{span_string}'.format(docid=mention.get('document_id'),
doce_or_kf_id=keyframe_id if keyframe_id else document_element_id,
span_string=span_string)
region = augment_mention_object(spanstring_to_object(self.logger, fq_span_string), self.get('document_mappings'), self.get('document_boundaries'))
if get_intersection_over_union(mention, region) > 0:
return True
return False
def contains_relaxed(self, mention, types, metatype):
document_id = mention.get('document_id')
document_element_id = mention.get('document_element_id')
keyframe_id = mention.get('keyframe_id')
doce_or_kf_id = keyframe_id if keyframe_id else document_element_id
for cluster_type in types:
top_level_type = get_top_level_type(cluster_type, metatype)
for key in self.get('regions'):
document_id_, doce_or_kf_id_, cluster_type_ = key.split(':')
metatype_ = self.get('ontology_type_mappings').get('type_to_metatype', cluster_type_)
top_level_type_ = get_top_level_type(cluster_type_, metatype_)
if document_id != document_id_: continue
if doce_or_kf_id != doce_or_kf_id_: continue
if top_level_type != top_level_type_: continue
for region_string in [s.__str__() for s in self.get('regions').get(key)]:
fq_region_string = '{docid}:{doce_or_kf_id}:{region_string}'.format(docid=document_id_,
doce_or_kf_id=doce_or_kf_id_,
region_string=region_string)
region = augment_mention_object(spanstring_to_object(self.logger, fq_region_string), self.get('document_mappings'), self.get('document_boundaries'))
if get_intersection_over_union(mention, region) > 0:
return True
return False
