def candidate_generation_recall_at()

in blink/candidate_retrieval/evaluator.py [0:0]

• 26 lines of code
• 9 McCabe index (conditional complexity)

    def candidate_generation_recall_at(self, ax=None, max_rank=None):
        processed_mentions = self.data
        total_num_of_docs = len(processed_mentions)

        gold_positions = np.array(
            [
                mention["gold_pos"]
                for mention in processed_mentions
                if mention["gold_pos"] >= 0
            ]
        )
        if ax == None:
            fig = plt.figure(figsize=(7, 7))
            ax = plt.subplot(111)
            ax.set_ylabel(str("Recall"))
            ax.set_xlabel(str("True entity rank"))

        rank_count_pairs = sorted(Counter(gold_positions).items(), key=lambda x: x[0])

        # rank_count_pairs = rank_count_pairs[:k]

        counts = [i[1] for i in rank_count_pairs]
        recall = np.cumsum(counts) / total_num_of_docs * 100
        rankings = [i[0] for i in rank_count_pairs]

        if max_rank is not None:
            for idx, rank in enumerate(rankings):
                if rank > max_rank:
                    rankings = rankings[:idx]
                    recall = recall[:idx]
                    break

        ax.plot(rankings, recall)