def validate_responses(args):
logger = Logger(args.log, args.log_specifications, sys.argv)
logger.record_event('DEFAULT_INFO', 'validation started')
document_mappings = DocumentMappings(
logger, args.parent_children, Encodings(logger, args.encodings),
CoreDocuments(logger, args.core_documents))
text_boundaries = TextBoundaries(logger, args.sentence_boundaries)
image_boundaries = ImageBoundaries(logger, args.image_boundaries)
video_boundaries = VideoBoundaries(logger, args.video_boundaries)
keyframe_boundaries = KeyFrameBoundaries(logger, args.keyframe_boundaries)
document_boundaries = {
'text': text_boundaries,
'image': image_boundaries,
'keyframe': keyframe_boundaries,
'video': video_boundaries
}
queries = TA3QuerySet(logger, args.queries) if args.queries else None
responses = ResponseSet(logger,
document_mappings,
document_boundaries,
args.input,
args.runid,
args.task,
queries=queries)
responses.write_valid_responses(args.output)
num_warnings, num_errors = logger.get_stats()
closing_message = 'validation finished (warnings:{}, errors:{})'.format(
num_warnings, num_errors)
logger.record_event('DEFAULT_INFO', closing_message)
print(closing_message)
if num_errors > 0:
exit(ERROR_EXIT_CODE)
exit(ALLOK_EXIT_CODE)
def generate_confidence_intervals(args):
logger = Logger(args.log, args.log_specifications, sys.argv)
if not args.input.endswith('.tab'):
logger.record_event('DEFAULT_CRITICAL_ERROR',
'input filename should be a *.tab.')
aggregate = {}
for element in args.aggregate.split(','):
key, value = element.split(':')
if key not in aggregate:
aggregate[key] = []
aggregate[key].append(value)
confidence_interval = ConfidenceIntervals(logger,
macro=args.macro,
input=args.input,
primary_key_col=args.primary_key,
score=args.score,
aggregate=aggregate,
document_id_col=args.document_id,
run_id_col=args.run_id,
sizes=args.sizes,
seed_value=args.seed)
output = {'pretty': args.pretty_output, 'tab': args.tab_output}
for output_format in output:
fh = open(output[output_format], 'w')
fh.write(confidence_interval.get('output', output_format))
fh.close()
exit(ALLOK_EXIT_CODE)
def align_clusters(args):
logger = Logger(args.log, args.log_specifications, sys.argv)
document_mappings = DocumentMappings(
logger, args.parent_children, Encodings(logger, args.encodings),
CoreDocuments(logger, args.core_documents))
text_boundaries = TextBoundaries(logger, args.sentence_boundaries)
image_boundaries = ImageBoundaries(logger, args.image_boundaries)
video_boundaries = VideoBoundaries(logger, args.video_boundaries)
keyframe_boundaries = KeyFrameBoundaries(logger, args.keyframe_boundaries)
document_boundaries = {
'text': text_boundaries,
'image': image_boundaries,
'keyframe': keyframe_boundaries,
'video': video_boundaries
}
annotated_regions = AnnotatedRegions(logger, document_mappings,
document_boundaries, args.regions)
os.mkdir(args.similarities)
os.mkdir(args.alignment)
for entry in sorted(os.scandir(args.gold), key=str):
if entry.is_dir() and entry.name.endswith('.ttl'):
kb = entry.name
message = 'aligning clusters in {}'.format(entry.name)
logger.record_event('DEFAULT_INFO', message)
print('At {}: {}'.format(
time.strftime("%m/%d/%Y %H:%M:%S", time.localtime()), message))
document_id = kb.replace('.ttl', '')
gold_mentions = '{}/{}/AIDA_P2_TA1_CM_A0001.rq.tsv'.format(
args.gold, kb)
gold_edges = '{}/{}/AIDA_P2_TA1_AM_A0001.rq.tsv'.format(
args.gold, kb)
system_mentions = '{}/{}/AIDA_P2_TA1_CM_A0001.rq.tsv'.format(
args.system, kb)
system_edges = '{}/{}/AIDA_P2_TA1_AM_A0001.rq.tsv'.format(
args.system, kb)
gold_mentions = gold_mentions if os.path.exists(
gold_mentions) else None
gold_edges = gold_edges if os.path.exists(gold_edges) else None
system_mentions = system_mentions if os.path.exists(
system_mentions) else None
system_edges = system_edges if os.path.exists(
system_edges) else None
similarities = '{}/{}.tab'.format(args.similarities, document_id)
alignment = '{}/{}.tab'.format(args.alignment, document_id)
check_for_paths_non_existance([similarities, alignment])
clusters = Clusters(logger, document_mappings, document_boundaries,
annotated_regions, gold_mentions, gold_edges,
system_mentions, system_edges)
clusters.print_similarities(similarities)
clusters.print_alignment(alignment)
exit(ALLOK_EXIT_CODE)
def filter_responses(args):
logger = Logger(args.log, args.log_specifications, sys.argv)
ontology_type_mappings = OntologyTypeMappings(logger,
args.ontology_type_mappings)
slot_mappings = SlotMappings(logger, args.slot_mappings)
document_mappings = DocumentMappings(
logger, args.parent_children, Encodings(logger, args.encodings),
CoreDocuments(logger, args.core_documents))
text_boundaries = TextBoundaries(logger, args.sentence_boundaries)
image_boundaries = ImageBoundaries(logger, args.image_boundaries)
video_boundaries = VideoBoundaries(logger, args.video_boundaries)
keyframe_boundaries = KeyFrameBoundaries(logger, args.keyframe_boundaries)
document_boundaries = {
'text': text_boundaries,
'image': image_boundaries,
'keyframe': keyframe_boundaries,
'video': video_boundaries
}
responses = ResponseSet(logger, ontology_type_mappings, slot_mappings,
document_mappings, document_boundaries, args.input,
args.runid)
annotated_regions = AnnotatedRegions(logger, document_mappings,
document_boundaries, args.regions)
run_filter_on_all_responses(responses, annotated_regions,
document_mappings, document_boundaries)
os.mkdir(args.output)
for input_filename in responses:
output_filename = input_filename.replace(responses.get('path'),
args.output)
dirname = os.path.dirname(output_filename)
if not os.path.exists(dirname):
os.mkdir(dirname)
output_fh = open(output_filename, 'w')
header_printed = False
for linenum in sorted(responses.get(input_filename), key=int):
entry = responses.get(input_filename).get(str(linenum))
if not header_printed:
output_fh.write('{}\n'.format(entry.get('header').get('line')))
header_printed = True
if not entry.get('valid'):
logger.record_event('EXPECTING_VALID_ENTRY',
entry.get('where'))
continue
if entry.get('passes_filter'):
output_fh.write(entry.__str__())
output_fh.close()
exit(ALLOK_EXIT_CODE)
def main(args):
logger = Logger(args.log, args.log_specifications, sys.argv)
type_mappings = Container(logger)
for entry in FileHandler(logger, args.ontology_type_mappings):
type_mappings.add(key=entry.get('full_type_ov'),
value=entry.get('full_type'))
text_boundaries = TextBoundaries(logger, args.sentence_boundaries)
image_boundaries = ImageBoundaries(logger, args.image_boundaries)
video_boundaries = VideoBoundaries(logger, args.video_boundaries)
keyframe_boundaries = KeyFrameBoundaries(logger, args.keyframe_boundaries)
document_boundaries = {
'text': text_boundaries,
'image': image_boundaries,
'keyframe': keyframe_boundaries,
'video': video_boundaries
}
output = []
for entry in FileHandler(logger, args.input):
document_id = entry.get('root_doc_id')
document_element_id = entry.get('doc_element_id')
modality = entry.get('media_type')
type = entry.get('type')
subtype = entry.get('subtype')
subsubtype = entry.get('subsubtype')
# apply patch to correct LDC's mistake in annotation
if type == 'personalsocial' and subtype == 'unspecified':
subtype = 'relationship'
full_type = '{type}.{subtype}.{subsubtype}'.format(
type=type, subtype=subtype, subsubtype=subsubtype)
full_type_cleaned = full_type.replace('.unspecified', '')
propercased_full_type = type_mappings.get(full_type_cleaned, None)
if propercased_full_type is None:
logger.record_event(
'DEFAULT_CRITICAL_ERROR',
'propercased_full_type is None for full_type: {}'.format(
full_type))
span_string = entry.get('span')
keyframe_id = None
keyframe_num = 0
if span_string == 'ENTIRE_DOCUMENT_ELEMENT':
document_boundary = document_boundaries.get(modality).get(
document_element_id)
span_string = document_boundary.__str__()
elif '-' in span_string:
start, end = span_string.split('-')
span_string = '({start},0)-({end},0)'.format(start=start, end=end)
elif '_' in span_string:
keyframe_id = span_string
keyframe_num = span_string.split('_')[1]
document_boundary = document_boundaries.get('keyframe').get(
keyframe_id)
span_string = document_boundary.__str__()
else:
span_string = None
output_object = {
'document_id': document_id,
'document_element_id': document_element_id,
'keyframe_id': keyframe_id,
'keyframe_num': int(keyframe_num),
'modality': modality,
'region': span_string,
'type': propercased_full_type,
}
output.append(output_object)
printed = {}
fh = open(args.output, 'w')
header = [
'document_id', 'document_element_or_keyframe_id', 'modality', 'region',
'type'
]
fh.write('{}\n'.format('\t'.join(header)))
for output_object in multisort(
output, (('document_id', False), ('modality', False),
('document_element_id', False), ('keyframe_num', False),
('region', False), ('type', False))):
line = get_line(output_object, header)
if line not in printed:
fh.write('{}\n'.format(line))
printed[line] = 1
fh.close()
exit(ALLOK_EXIT_CODE)