def evaluate_example()

in src/lighteval/metrics/imports/data_stats_metric.py [0:0]

• 31 lines of code
• 8 McCabe index (conditional complexity)

    def evaluate_example(self, summary, input_text):
        if self.tokenize:
            input_text = _en(input_text, disable=["tagger", "parser", "ner", "textcat"])
            input_text = [tok.text for tok in input_text]
            summary = _en(summary, disable=["tagger", "parser", "ner", "textcat"])
            summary = [tok.text for tok in summary]
        fragments = Fragments(summary, input_text, case=self.case)
        coverage = fragments.coverage()
        density = fragments.density()
        compression = fragments.compression()
        score_dict = {"coverage": coverage, "density": density, "compression": compression}
        tokenized_summary = fragments.summary
        tokenized_text = fragments.text
        score_dict["summary_length"] = len(tokenized_summary)
        for i in range(1, self.n_gram + 1):
            input_ngrams = list(find_ngrams(tokenized_text, i))
            summ_ngrams = list(find_ngrams(tokenized_summary, i))
            input_ngrams_set = set(input_ngrams)
            summ_ngrams_set = set(summ_ngrams)
            intersect = summ_ngrams_set.intersection(input_ngrams_set)
            try:
                score_dict[f"percentage_novel_{i}-gram"] = (len(summ_ngrams_set) - len(intersect)) / float(
                    len(summ_ngrams_set)
                )
                ngramCounter = Counter()
                ngramCounter.update(summ_ngrams)
                repeated = [key for key, val in ngramCounter.items() if val > 1]
                score_dict[f"percentage_repeated_{i}-gram_in_summ"] = len(repeated) / float(len(summ_ngrams_set))
            except ZeroDivisionError:
                continue
        return score_dict