def t2t_bleu(targets, predictions):
"""Tokenizes with the bleu_tokenize method from the t2t library then
calls the compute_bleu function
Args:
targets: a list of strings, the target from the validation set
preditcions: a list of strings, the model predictions
Returns:
a dictionary: {"t2t_bleu": bleu_value}
"""
targets_tokens = [bleu_hook.bleu_tokenize(x) for x in targets]
predictions_tokens = [bleu_hook.bleu_tokenize(x) for x in predictions]
return {"t2t_bleu": 100 * bleu_hook.compute_bleu(targets_tokens,
predictions_tokens)}
def testBleuTokenize(self):
self.assertEqual(bleu_hook.bleu_tokenize(u"hi, “there”"),
[u"hi", u",", u"“", u"there", u"”"])
def testBleuTokenize(self):
self.assertEqual(bleu_hook.bleu_tokenize(u"hi, “there”"),
[u"hi", u",", u"“", u"there", u"”"])
def testBleuTokenize(self): self.assertEqual(bleu_hook.bleu_tokenize(u'hi, “there”'), [u'hi', u',', u'“', u'there', u'”'])
