def score_manifold_model()

in archs/models.py [0:0]

• 13 lines of code
• 4 McCabe index (conditional complexity)

    def score_manifold_model(self, scores, obj_truth, bias=0.0):
        # put everything on CPU
        scores = {k: v.cpu() for k, v in scores.items()}
        obj_truth = obj_truth.cpu()
        # gather scores for all relevant (a,o) pairs
        scores = torch.stack(
            [scores[(attr, obj)] for attr, obj in self.dset.pairs],
            1)  # (B, #pairs)
        orig_scores = scores.clone()
        mask = self.seen_mask.repeat(scores.shape[0], 1)
        scores[1 - mask] += bias
        results = self.generate_predictions(scores, obj_truth)
        results['biased_scores'] = scores
        results['scores'] = orig_scores
        return results