def learn_new_triples(predicate):
"""learns new triples and stores them in a file"""
sc = get_sentence_classifier(predicate)
if predicate in numeric_predicates:
pattern = '<%s> <%s> "%s"^^<http://www.w3.org/2001/XMLSchema#int> .'
else:
pattern = '<%s> <%s> "%s"@pl .'
entities = CandidatesSelector.get_candidates(predicate)
entities = entities[:candidates_limit]
if verbose:
print '%s candidates identified' % len(entities)
n = 1000
entities_list = [
entities[i * n:(i + 1) * n]
for i in xrange(int(ceil(len(entities) / n)))
]
out = open(results_path + 'triples-%s' % predicate, 'w')
ve = ValueExtractor(predicate, sc.extractor_training_data)
for entities in entities_list:
extracted_sentences = sc.extract_sentences(entities)
values = ve.extract_values(extracted_sentences)
for e, v in values.iteritems():
if v:
print >> out, pattern % (
full_resource_name(e).encode('utf-8'),
full_predicate_name(predicate).encode('utf-8'), v)
def run_evaluation(predicate, sentence_limit=None):
entities, true_values = get_test_data(predicate)
sc = get_sentence_classifier(predicate, sentence_limit)
true_values = dict((k, v) for k, v in true_values.iteritems() if k in entities)
if verbose:
print '%d entities were used in evaluation.' % len(entities)
extracted_sentences = sc.extract_sentences(entities)
ve = ValueExtractor(predicate, sc.extractor_training_data)
values = ve.extract_values(extracted_sentences)
print '%s results:' % predicate
stats, fp, fn = ValueExtractorEvaluator.evaluate(true_values, values)
table_format = '%30s %30s %20s %10s'
print 'Error table:'
print table_format % ('Subject:', 'Gold standard values:', 'Extracted value:', 'Error:')
for entity, value in values.iteritems():
if entity not in true_values:
true_values[entity] = '-'
for entity, true_value in true_values.iteritems():
if entity in fp and entity in fn:
err = 'FP/FN'
elif entity in fp:
err = 'FP'
elif entity in fn:
err = 'FN'
else:
err = ''
print table_format % (entity[:30], ', '.join(true_value), values[entity] if entity in values else '-', err)
print '\n\n'
return stats
def learn_new_triples(predicate):
"""learns new triples and stores them in a file"""
sc = get_sentence_classifier(predicate)
if predicate in numeric_predicates:
pattern = '<%s> <%s> "%s"^^<http://www.w3.org/2001/XMLSchema#int> .'
else:
pattern = '<%s> <%s> "%s"@pl .'
entities = CandidatesSelector.get_candidates(predicate)
entities = entities[: candidates_limit]
if verbose:
print '%s candidates identified' % len(entities)
n = 1000
entities_list = [
entities[i*n : (i+1)*n]
for i in xrange(int(ceil(len(entities) / n)))
]
out = open(results_path + 'triples-%s' % predicate, 'w')
ve = ValueExtractor(predicate, sc.extractor_training_data)
for entities in entities_list:
extracted_sentences = sc.extract_sentences(entities)
values = ve.extract_values(extracted_sentences)
for e, v in values.iteritems():
if v:
print >>out, pattern % (
full_resource_name(e).encode('utf-8'),
full_predicate_name(predicate).encode('utf-8'),
v
)
