def score(self, target, prediction):
"""Calculates rouge scores between the target and prediction.
Args:
target: Text containing the target (ground truth) text.
prediction: Text containing the predicted text.
Returns:
A dict mapping each rouge type to a Score object.
Raises:
ValueError: If an invalid rouge type is encountered.
"""
target_tokens = tokenize.tokenize(target, self._stemmer)
prediction_tokens = tokenize.tokenize(prediction, self._stemmer)
result = {}
for rouge_type in self.rouge_types:
if rouge_type == "rougeL":
# Rouge from longest common subsequences.
scores = _score_lcs(target_tokens, prediction_tokens)
elif re.match(r"rouge[0-9]$", rouge_type):
# Rouge from n-grams.
n = int(rouge_type[5:])
if n <= 0:
raise ValueError("rougen requires positive n: %s" %
rouge_type)
target_ngrams = _create_ngrams(target_tokens, n)
prediction_ngrams = _create_ngrams(prediction_tokens, n)
scores = _score_ngrams(target_ngrams, prediction_ngrams)
else:
raise ValueError("Invalid rouge type: %s" % rouge_type)
result[rouge_type] = scores
return result
def score(self, target, prediction):
"""Calculates rouge scores between the target and prediction.
Args:
target: Text containing the target (ground truth) text.
prediction: Text containing the predicted text.
Returns:
A dict mapping each rouge type to a Score object.
Raises:
ValueError: If an invalid rouge type is encountered.
"""
# Pre-compute target tokens and prediction tokens for use by different
# types, except if only "rougeLsum" is requested.
if len(self.rouge_types) == 1 and self.rouge_types[0] == "rougeLsum":
target_tokens = None
prediction_tokens = None
else:
target_tokens = tokenize.tokenize(target, self._stemmer)
prediction_tokens = tokenize.tokenize(prediction, self._stemmer)
result = {}
for rouge_type in self.rouge_types:
if rouge_type == "rougeL":
# Rouge from longest common subsequences.
scores = _score_lcs(target_tokens, prediction_tokens)
elif rouge_type == "rougeLsum":
# Note: Does not support multi-line text.
def get_sents(text):
# Assume sentences are separated by newline.
sents = six.ensure_str(text).split("\n")
sents = [x for x in sents if len(x)]
return sents
target_tokens_list = [
tokenize.tokenize(s, self._stemmer) for s in get_sents(target)]
prediction_tokens_list = [
tokenize.tokenize(s, self._stemmer) for s in get_sents(prediction)]
scores = _summary_level_lcs(target_tokens_list,
prediction_tokens_list)
elif re.match(r"rouge[0-9]$", six.ensure_str(rouge_type)):
# Rouge from n-grams.
n = int(rouge_type[5:])
if n <= 0:
raise ValueError("rougen requires positive n: %s" % rouge_type)
target_ngrams = _create_ngrams(target_tokens, n)
prediction_ngrams = _create_ngrams(prediction_tokens, n)
scores = _score_ngrams(target_ngrams, prediction_ngrams)
else:
raise ValueError("Invalid rouge type: %s" % rouge_type)
result[rouge_type] = scores
return result
def test_give_me_a_name(self):
self.assertEqual(['one', 'two', 'three'],
tokenize.tokenize('one Two three', None))
self.assertEqual(['one', 'two', 'three'],
tokenize.tokenize('one\n Two \nthree', None))
def tokenize(self, text):
return tokenize.tokenize(text, self._stemmer)