def number_feature(data_set_path: str, db_path: str, max_sent_num: int):
from common.dataset.reader import JSONLineReader
db = FeverDocDB(db_path)
jlr = JSONLineReader()
lines = jlr.read(data_set_path)
num_feat = np.zeros([len(lines), max_sent_num, 3], dtype=np.int32)
for i, line in enumerate(lines):
claim_text = line['claim']
claim_tokens = tokenize(claim_text)
all_nums = set()
for token in claim_tokens:
if is_token_numeric(token):
all_nums.add(float(token))
for j, evidence in enumerate(line['predicted_evidence']):
if j >= max_sent_num:
break
page, line_num = evidence[-2], evidence[-1]
all_evidence_nums = []
evidence_text = evidence_num_to_text(db, page, line_num)
evidence_tokens = tokenize(evidence_text)
for token in evidence_tokens:
if is_token_numeric(token):
all_evidence_nums.append(float(token))
has_num = len(all_evidence_nums) > 0
has_identical_num = any(n in all_nums for n in all_evidence_nums)
has_different_num = any(n not in all_nums
for n in all_evidence_nums)
num_feat[i][j][0], num_feat[i][j][1], num_feat[i][j][
2] = _interprete_num_result(has_num, has_identical_num,
has_different_num)
return num_feat
def single_sentence_set_2_ids_given_vocab(texts, vocab_dict):
logger = LogHelper.get_logger("single_sentence_set_2_ids_given_vocab")
doc_ids = []
out_of_vocab_counts = 0
for sent in texts:
tokens = tokenize(sent)
word_ids = []
for token in tokens:
if token.lower() in vocab_dict:
word_ids.append(vocab_dict[token.lower()])
else:
out_of_vocab_counts += 1
word_ids.append(vocab_dict['UNK'])
doc_ids.append(word_ids)
logger.debug("{} times out of vocab".format(str(out_of_vocab_counts)))
return doc_ids
def single_sentence_set_2_fasttext_embedded(sents: List[str],
fasttext_model: Union[str,
FastText]):
logger = LogHelper.get_logger("single_sentence_set_2_fasttext_embedded")
if type(fasttext_model) == str:
fasttext_model = FastText.load_fasttext_format(fasttext_model)
fasttext_embeddings = []
for sent in sents:
tokens = tokenize(sent)
sent_embeddings = []
for token in tokens:
try:
sent_embeddings.append(fasttext_model[token.lower()])
except KeyError:
sent_embeddings.append(np.ones([dim_fasttext], np.float32))
fasttext_embeddings.append(sent_embeddings)
return fasttext_embeddings, fasttext_model
voc_dict['UNK'] = 1
return voc_dict
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('db', help='/path/to/db/file')
parser.add_argument('output', help='/path/to/output/pickle/file')
args = parser.parse_args()
LogHelper.setup()
logger = LogHelper.get_logger("generate_vocab_all_wiki")
db = FeverDocDB(args.db)
vocab = set()
for doc in tqdm(db.get_doc_ids()):
lines = db.get_doc_lines(doc)
lines = lines.split("\n")
for line in lines:
segments = line.split("\t")
if len(segments) < 2:
continue
line = segments[1]
if line.strip() == "":
continue
tokens = set(token.lower() for token in tokenize(clean_text(line)))
vocab.update(tokens)
logger.info("total size of vocab: " + str(len(vocab)))
vocab_dict = vocab_map(vocab)
del vocab
with open(args.output, 'wb') as f:
pickle.dump(vocab_dict, f, protocol=pickle.HIGHEST_PROTOCOL)
