44d2847610944f56a06b7cfa54faadb66e130a83,allennlp/training/metrics/categorical_accuracy.py,CategoricalAccuracy,__call__,#CategoricalAccuracy#Any#Any#Any#,31

Before Change

            if self._top_k == 1:
                top_k = predictions.max(-1)[1].unsqueeze(-1)
            else:
                top_k = predictions.topk(min(self._top_k, predictions.shape[-1]), -1)[1]

            // This is of shape (batch_size, ..., top_k).
            correct = top_k.eq(gold_labels.unsqueeze(-1)).float()
        else:

After Change

            self.total_count += gold_labels.numel()
        self.correct_count += correct.sum()

        if is_distributed():
            _correct_count = torch.tensor(self.correct_count).to(device)
            _total_count = torch.tensor(self.total_count).to(device)
            dist.all_reduce(_correct_count, op=dist.ReduceOp.SUM)
            dist.all_reduce(_total_count, op=dist.ReduceOp.SUM)
            self.correct_count = _correct_count.item()
            self.total_count = _total_count.item()

    def get_metric(self, reset: bool = False):
        
        // Returns

Instances