def recategorize_file(self, file_path):
for i, amr in enumerate(AMRIO.read(file_path), 1):
self.recategorize_graph(amr)
yield amr
if i % 1000 == 0:
logger.info('Processed {} examples.'.format(i))
logger.info('Done.\n')
def _read(self, file_path):
# if `file_path` is a URL, redirect to the cache
file_path = cached_path(file_path)
logger.info("Reading instances from lines in file at: %s", file_path)
for amr in AMRIO.read(file_path):
yield self.text_to_instance(amr)
self.report_coverage()
def _get_senseless_node_counter(amr_train_files):
logger.info('Building the senseless node counter.')
sense_less_nodes = []
for amr_file in amr_train_files:
for amr in AMRIO.read(amr_file):
for node in amr.graph.get_nodes():
for attr, value in node.get_senseless_attributes():
sense_less_nodes.append(value)
return Counter(sense_less_nodes)
def _update_counter_from_train_files(self, amr_train_files, base_freq=1):
logger.info('Updating (lemma, frame) counter from AMR train files.')
for file_path in amr_train_files:
for amr in AMRIO.read(file_path):
for node in amr.graph.get_nodes():
for _, frame in node.get_frame_attributes():
frame_lemma = re.sub(WORDSENSE_RE, '', frame)
self._update_counter(self.lemma_frame_counter, frame_lemma, frame, base_freq)
self._update_counter(self.frame_lemma_counter, frame, frame_lemma, base_freq)
def dump_spotlight_wiki(self, file_path):
sent_map = {}
for i, amr in enumerate(AMRIO.read(file_path), 1):
if i % 20 == 0:
print('+', end='')
sent = amr.sentence
wiki = self.spotlight_wiki(sent)
sent_map[sent] = wiki
sleep(0.1)
with open(os.path.join(self.util_dir, 'spotlight_wiki.json'),
'w',
encoding='utf-8') as f:
json.dump(sent_map, f)
def dump_spotlight_wiki(self, amr_files):
#!!! This function has been changed by Deng Cai
sent_map = {}
for file_path in amr_files:
for i, amr in enumerate(AMRIO.read(file_path), 1):
if i % 20 == 0:
print('+', end='')
sent = amr.sentence
wiki = self.spotlight_wiki(sent)
sent_map[sent] = wiki
sleep(0.1)
with open(os.path.join(self.util_dir, 'spotlight_wiki.json'),
'w',
encoding='utf-8') as f:
json.dump(sent_map, f)
def create_dependency_parser_feature_from_file(annotator, filepath):
dependency_feature_data = []
amrs = []
sentence_ids = []
with open(filepath + '.features', 'w', encoding='utf-8') as f:
for i, amr in enumerate(AMRIO.read(filepath), 1):
if i % 100 == 0:
print('{} processed.'.format(i))
annotation = annotator.annotate(amr.sentence)
dump_amr_features(amr, annotation, f)
sentence_data = create_dependency_parser_feature(annotation, amr.sentence, i)
dependency_feature_data.append(sentence_data)
sentence_ids.append(i)
amrs.append(amr)
dataset_dict = {
'sentence_id': sum([sum([sentence_data['sentence_id'] for sentence_data in dependency_feature_data], [])],[]),
'sequence': sum([sum([sentence_data['sequence'] for sentence_data in dependency_feature_data], [])],[]),
'parent': sum([sum([sentence_data['parent'] for sentence_data in dependency_feature_data], [])],[]),
'parent_position': sum([sum([sentence_data['parent_position'] for sentence_data in dependency_feature_data], [])],[]),
'child': sum([sum([sentence_data['child'] for sentence_data in dependency_feature_data], [])],[]),
'child_position': sum([sum([sentence_data['child_position'] for sentence_data in dependency_feature_data], [])],[]),
'is_root' : sum([sum([sentence_data['is_root'] for sentence_data in dependency_feature_data], [])],[]),
'parent_ner': sum([sum([sentence_data['parent_ner'] for sentence_data in dependency_feature_data], [])],[]),
'child_ner': sum([sum([sentence_data['child_ner'] for sentence_data in dependency_feature_data], [])],[]),
'parent_pos': sum([sum([sentence_data['parent_pos'] for sentence_data in dependency_feature_data], [])],[]),
'dependency_role': sum([sum([sentence_data['dependency_role'] for sentence_data in dependency_feature_data], [])],[]),
'child_pos': sum([sum([sentence_data['child_pos'] for sentence_data in dependency_feature_data], [])],[])
}
amr_dict = {
'sentence_id': sentence_ids,
'amr': [str(amr.graph) for amr in amrs]
}
dependency_feature_df = pd.DataFrame(dataset_dict)
amr_df = pd.DataFrame(amr_dict)
return dependency_feature_df, amr_df
def load_data(self):
logger.info("Parsing and linearizing the AMR dataset")
train_amr = AMRIO.read(self.train_file)
for i, amr in tqdm(enumerate(train_amr), desc='Train AMR'):
# Raw version
if self.small and i > 50:
break
raw_amr = []
for amr_line in str(amr.graph).splitlines():
striped_amr = amr_line.strip()
raw_amr.append(striped_amr)
self.X_train_raw.append(" ".join(raw_amr))
linearized_amr = self.get_list(amr)
self.X_train.append(linearized_amr[1:])
self.Y_train.append(amr.sentence)
self.Y_train_tok.append(amr.tokens)
# Vocabulary Create dictionaries and simplify list
simpl = list()
simpl_only_nodes = list()
for step in linearized_amr:
if step not in self.lin_to_int.keys():
self.lin_to_int[step] = len(self.lin_to_int)
self.int_to_lin[len(self.int_to_lin)] = step
# simplyfied AMR version
step, edge = self.simplify(step)
simpl.append(step)
if not step.startswith(":"):
simpl_only_nodes.append(step)
# Identify edges and save them
if edge and step not in self.edges:
self.edges.append(step)
self.X_train_simple.append(simpl)
self.X_train_simple_only_nodes.append(simpl_only_nodes)
sent = amr.sentence.split()
for word in sent:
if word not in self.word_to_int.keys():
self.word_to_int[word] = len(self.word_to_int)
self.int_to_word[len(self.int_to_word)] = word
if self.use_silver_data:
print("Processing silver data from", self.silver_train_file)
ii = 0
silver_train_amr = AMRIO.read(self.silver_train_file)
for i, amr in enumerate(silver_train_amr):
if self.small and i > 50:
break
# Raw version
raw_amr = []
ii += 1
linearized_amr = self.get_list(amr)
if linearized_amr is None:
continue
for amr_line in str(amr.graph).splitlines():
striped_amr = amr_line.strip()
raw_amr.append(striped_amr)
self.X_silver_train_raw.append(" ".join(raw_amr))
self.X_silver_train.append(linearized_amr[1:])
self.Y_silver_train.append(amr.sentence)
self.Y_silver_train_tok.append(amr.tokens)
# Vocabulary Create dictionaries and simplify list
simpl = list()
simpl_only_nodes = list()
for step in linearized_amr:
if step not in self.lin_to_int.keys():
self.lin_to_int[step] = len(self.lin_to_int)
self.int_to_lin[len(self.int_to_lin)] = step
# simplyfied AMR version
step, edge = self.simplify(step)
simpl.append(step)
if not step.startswith(":"):
simpl_only_nodes.append(step)
# Identify edges and save them
if edge and step not in self.edges:
self.edges.append(step)
self.X_silver_train_simple.append(simpl)
self.X_silver_train_simple_only_nodes.append(simpl_only_nodes)
sent = amr.sentence.split()
for word in sent:
if word not in self.word_to_int.keys():
self.word_to_int[word] = len(self.word_to_int)
self.int_to_word[len(self.int_to_word)] = word
print("Silver data with size:", len(self.X_silver_train_raw))
else:
print("No silver data performed")
dev_amr = AMRIO.read(self.dev_file)
for i, amr in tqdm(enumerate(dev_amr), desc='Dev AMR'):
if self.small and i > 50:
break
# Raw input
raw_amr = []
for amr_line in str(amr.graph).splitlines():
striped_amr = amr_line.strip()
raw_amr.append(striped_amr)
self.X_dev_raw.append(" ".join(raw_amr))
linearized_amr = self.get_list(amr)
self.X_dev.append(linearized_amr[1:])
self.Y_dev.append(amr.sentence)
self.Y_dev_tok.append(amr.tokens)
# simplyfied AMR version
simpl = list()
simpl_only_nodes = list()
for step in linearized_amr:
step, edge = self.simplify(step)
simpl.append(step)
if not step.startswith(":"):
simpl_only_nodes.append(step)
if edge and step not in self.edges:
self.edges.append(step)
self.X_dev_simple.append(simpl)
self.X_dev_simple_only_nodes.append(simpl_only_nodes)
test_amr = AMRIO.read(self.test_file)
self.amr_test = test_amr
for i, amr in tqdm(enumerate(test_amr), desc='Test AMR'):
if self.small and i > 50:
break
# Raw version
raw_amr = []
for amr_line in str(amr.graph).splitlines():
striped_amr = amr_line.strip()
raw_amr.append(striped_amr)
self.X_test_raw.append(" ".join(raw_amr))
linearized_amr = self.get_list(amr)
self.X_test.append(linearized_amr[1:])
self.Y_test.append(amr.sentence)
self.Y_test_tok.append(amr.tokens)
# simplyfied AMR version
simpl = list()
simpl_only_nodes = list()
for step in linearized_amr:
step, edge = self.simplify(step)
simpl.append(step)
if not step.startswith(":"):
simpl_only_nodes.append(step)
if edge and step not in self.edges:
self.edges.append(step)
self.X_test_simple.append(simpl)
self.X_test_simple_only_nodes.append(simpl_only_nodes)
def expand_file(self, file_path):
for i, amr in enumerate(AMRIO.read(file_path)):
self.expand_graph(amr)
yield amr
self.print_stats()
if __name__ == '__main__':
import argparse
from stog.data.dataset_readers.amr_parsing.io import AMRIO
parser = argparse.ArgumentParser('feature_annotator.py')
parser.add_argument('files', nargs='+', help='files to annotate.')
parser.add_argument('--compound_file', default='')
args = parser.parse_args()
annotator = FeatureAnnotator('http://localhost:9000', args.compound_file)
for file_path in args.files:
logger.info('Processing {}'.format(file_path))
for i, amr in enumerate(AMRIO.read(file_path), 1):
if i % 1000 == 0:
logger.info('{} processed.'.format(i))
annotation = annotator(amr.sentence)
amr.tokens = annotation['tokens']
amr.lemmas = annotation['lemmas']
amr.pos_tags = annotation['pos_tags']
amr.ner_tags = annotation['ner_tags']
amr.original = annotation['original']
with open(file_path + '.features', 'a', encoding='utf-8') as f:
AMRIO.dump([amr], f)
logger.info('Done!')
parser = argparse.ArgumentParser("text_anonymizor.py")
parser.add_argument('--amr_file', nargs="+", required=True, help="File to anonymize.")
parser.add_argument('--util_dir')
parser.add_argument('--lang')
parser.add_argument('--exclude_ners', action="store_true", help="consider NER tags for entities not found in training.")
args = parser.parse_args()
if args.lang=="en":
text_anonymizor = TextAnonymizor.from_json(os.path.join(args.util_dir,
"text_anonymization_rules.json"))
lang_stopwords=None
lang2en_span=None
lang2en_bn=None
else:
text_anonymizor = TextAnonymizor.from_json(os.path.join(args.util_dir,"text_anonymization_en-{}.json".format(args.lang)))
lang_stopwords = set([x.rstrip() for x in open("data/cross-lingual-babelnet_mappings/stopwords_{}.txt".format(args.lang))])
lang2en_span=load_name_span_map("data/cross-lingual-babelnet_mappings/name_span_en_{}_map_amr_bn.json".format(args.lang), args.lang)
lang2en_bn=load_name_bn_wiki_map("data/cross-lingual-babelnet_mappings/namedEntity_wiki_synsets.{}.tsv".format(args.lang.upper()))
for amr_file in args.amr_file:
with open(amr_file + ".recategorize{}".format("_noner" if args.exclude_ners else ""), "w", encoding="utf-8") as f:
for amr in tqdm(AMRIO.read(amr_file, lang=args.lang)):
amr.abstract_map = text_anonymizor(amr)
f.write(str(amr) + "\n\n")
def wikify_file(self, file_path):
for i, amr in enumerate(AMRIO.read(file_path)):
self.wikify_graph(amr)
yield amr
return cls(
text_maps=d["text_maps"],
priority_lists=d["priority_lists"],
_VNE=d["VNE"],
_LOCEN1=d["LOCEN1"],
_LOCEN2=d["LOCEN2"],
_N=d["N"],
_M=d["M"],
_R=d["R"],
_INVP=d["INVP"],
_INVS=d["INVS"],
)
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser("text_anonymizor.py")
parser.add_argument('--amr_file', required=True, help="File to anonymize.")
parser.add_argument('--util_dir')
args = parser.parse_args()
text_anonymizor = TextAnonymizor.from_json(
os.path.join(args.util_dir, "text_anonymization_rules.json"))
with open(args.amr_file + ".recategorize", "w", encoding="utf-8") as f:
for amr in AMRIO.read(args.amr_file):
amr.abstract_map = text_anonymizor(amr)
f.write(str(amr) + "\n\n")
if token == '911':
index = i
break
else:
break
amr.replace_span([index], ['09', '11'], ['CD', 'CD'], ['DATE', 'DATE'])
def replace_NT_dollar_abbr(amr):
# Replace 'NT' in front of '$' with 'Taiwan'.
for i, token in enumerate(amr.tokens):
if token == 'NT' and len(amr.tokens) > i + 1 and amr.tokens[i + 1] in (
'$', 'dollars', 'dollar'):
amr.replace_span([i], ['Taiwan'], ['NNP'], ['COUNTRY'])
if __name__ == '__main__':
import argparse
from stog.data.dataset_readers.amr_parsing.io import AMRIO
parser = argparse.ArgumentParser('input_cleaner.py')
parser.add_argument('--amr_files', nargs='+', default=[])
args = parser.parse_args()
for file_path in args.amr_files:
with open(file_path + '.input_clean', 'w', encoding='utf-8') as f:
for amr in AMRIO.read(file_path):
clean(amr)
f.write(str(amr) + '\n\n')
def read(self, file_path):
for amr in AMRIO.read(file_path):
yield self(amr)
def dump_amr_features(amr, annotation, f):
amr.tokens = annotation['tokens']
amr.lemmas = annotation['lemmas']
amr.pos_tags = annotation['pos_tags']
amr.ner_tags = annotation['ner_tags']
AMRIO.dump([amr], f)
