Example #1
0
    def score_sentence(self, ref, out):
        if self.case_insensitive:
            ref = corpus_utils.lower(ref)
            out = corpus_utils.lower(out)

        if self._stemmer:
            ref = [self._stemmer.stem(x) if len(x) > 3 else x for x in ref]
            out = [self._stemmer.stem(x) if len(x) > 3 else x for x in out]

        if self.rouge_type == 'rougeL':
            scores = rouge_scorer._score_lcs(ref, out)
        elif re.match(r"rouge[0-9]$", self.rouge_type):
            n = int(self.rouge_type[5:])
            if n <= 0:
                raise ValueError(
                    f"rougen requires positive n: {self.rouge_type}")
            ref_ngrams = rouge_scorer._create_ngrams(ref, n)
            out_ngrams = rouge_scorer._create_ngrams(out, n)
            scores = rouge_scorer._score_ngrams(ref_ngrams, out_ngrams)
        else:
            raise ValueError(f"Invalid rouge type: {self.rouge_type}")

        if self.score_type == 'fmeasure':
            return scores.fmeasure, None
        elif self.score_type == 'precision':
            return scores.precision, None
        elif self.score_type == 'recall':
            return scores.recall, None
        else:
            raise ValueError(f"Invalid score type: {self.score_type}")
Example #2
0
    def score_sentence(self, ref, out, src=None):
        if self.case_insensitive:
            ref = corpus_utils.lower(ref)
            out = corpus_utils.lower(out)

        if self._stemmer:
            ref = [self._stemmer.stem(x) if len(x) > 3 else x for x in ref]
            out = [self._stemmer.stem(x) if len(x) > 3 else x for x in out]

        if self.rouge_type == 'rougeL':
            ref, out = self.tokenize(" ".join(ref)), self.tokenize(
                " ".join(out))
            scores = rouge_scorer._score_lcs(ref, out)
        elif self.rouge_type == 'rougeLsum':
            refs = [self.tokenize(s) for s in self.get_sents(ref)]
            outs = [self.tokenize(s) for s in self.get_sents(out)]
            scores = rouge_scorer._summary_level_lcs(refs, outs)
        elif re.match(r"rouge[0-9]$", self.rouge_type):
            ref, out = self.tokenize(" ".join(ref)), self.tokenize(
                " ".join(out))
            n = int(self.rouge_type[5:])
            if n <= 0:
                raise ValueError(
                    f"rougen requires positive n: {self.rouge_type}")
            ref_ngrams = rouge_scorer._create_ngrams(ref, n)
            out_ngrams = rouge_scorer._create_ngrams(out, n)
            scores = rouge_scorer._score_ngrams(ref_ngrams, out_ngrams)
        else:
            raise ValueError(f"Invalid rouge type: {self.rouge_type}")

        if self.score_type == 'fmeasure':
            score_value = scores.fmeasure
        elif self.score_type == 'precision':
            score_value = scores.precision
        elif self.score_type == 'recall':
            score_value = scores.recall
        else:
            raise ValueError(f"Invalid score type: {self.score_type}")

        return self.scale * score_value, None