rouge.py 文件源码

def compute_score(self, gts, res):
        """
        Computes Rouge-L score given a set of reference and candidate sentences for the dataset
        Invoked by evaluate_captions.py

        :param hypo_for_image: dict : candidate / test sentences with "image name" key and "tokenized sentences" as values
        :param ref_for_image: dict : reference MS-COCO sentences with "image name" key and "tokenized sentences" as values
        :returns: average_score: float (mean ROUGE-L score computed by averaging scores for all the images)
        """
        assert(gts.keys() == res.keys())
        imgIds = gts.keys()

        score = []
        for id in imgIds:
            hypo = res[id]
            ref  = gts[id]

            score.append(self.calc_score(hypo, ref))

            # Sanity check.
            assert(type(hypo) is list)
            assert(len(hypo) == 1)
            assert(type(ref) is list)
            assert(len(ref) > 0)

        average_score = np.mean(np.array(score))
        return average_score, np.array(score)