def test_epoch_end()

in run_ranking.py [0:0]

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

    def test_epoch_end(self, outputs):
        logits = torch.cat([x['logits'] for x in outputs], dim=0).cpu()
        scores = F.sigmoid(logits).squeeze(1).numpy().tolist() # (num_test, )

        lines = []
        count = 0
        with jsonlines.open(os.path.join(self.hparams.data_dir, f"{self.hparams.question_type}.dev.es_retrieved.processed.jsonl")) as f:
            for line in f.iter():
                line['passages'] = line['passages'][:100]
                line['tables'] = line['tables'][:100]

                for cand in line['passages']:
                    cand['rank_score'] = scores[count]
                    count += 1
                    if count % 10000 == 0:
                        logger.info(f"Finished {count} lines")
                for cand in line['tables']:
                    cand['rank_score'] = scores[count]
                    count += 1
                    if count % 10000 == 0:
                        logger.info(f"Finished {count} lines")
                lines.append(line)

        score_dir = os.path.join(self.hparams.data_dir, "scores")
        if not os.path.exists(score_dir):
            os.makedirs(score_dir)
        with jsonlines.open(os.path.join(score_dir, f"{self.hparams.question_type}.dev.es_retrieved.processed.scores.jsonl"), 'w') as writer:
            writer.write_all(lines)

        return {'avg_test_loss': -1}