File size: 902 Bytes
4f77642 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 | import pickle
import string
import fire
import nltk
def main(target_lang='english', data_path='data/data.pkl'):
key_hand = 'txt_en' if target_lang == 'english' else 'txt_ru'
key_auto = 'txt_ru2en' if target_lang == 'english' else 'txt_en2ru'
tokenizer_en = nltk.tokenize.TweetTokenizer(preserve_case=False)
data = pickle.load(open(data_path, 'rb'))
references = []
hypotheses = []
for doc in data:
references += [
[[tok for tok in tokenizer_en.tokenize(vars(edu).get(key_hand)) if not tok in string.punctuation]]
for edu in doc.edus]
hypotheses += [[tok for tok in tokenizer_en.tokenize(vars(edu).get(key_auto)) if not tok in string.punctuation]
for edu in doc.edus]
print('BLEU score =', nltk.translate.bleu_score.corpus_bleu(references, hypotheses))
if __name__ == '__main__':
fire.Fire(main)
|