9766d2c13ddca10d5e45280450ee9cab649afb18,pytorch/cls_metrics.py,TopKAccuracy,update,#TopKAccuracy#Any#Any#,105

Before Change

            pred_label = preds.cpu().numpy().astype(np.int32)
            pred_label = np.argpartition(pred_label, -self.top_k)
            label = labels.cpu().numpy().astype(np.int32)
            assert (label.shape == pred_label.shape)
            num_samples = pred_label.shape[0]
            num_dims = len(pred_label.shape)
            if num_dims == 1:
                self.sum_metric += (pred_label.flat == label.flat).sum()

After Change

        
        assert (len(labels) == len(preds))
        with torch.no_grad():
            if self.torch_like:
                _, pred = preds.topk(k=self.top_k, dim=1, largest=True, sorted=True)
                pred = pred.t()
                correct = pred.eq(labels.view(1, -1).expand_as(pred))
                num_correct = correct.view(-1).float().sum(dim=0, keepdim=True).item()
                num_samples = labels.size(0)
                assert (num_correct <= num_samples)
                self.sum_metric += num_correct
                self.global_sum_metric += num_correct
                self.num_inst += num_samples
                self.global_num_inst += num_samples
            else:
                assert(len(preds.shape) <= 2), "Predictions should be no more than 2 dims"
                pred_label = preds.cpu().numpy().astype(np.int32)
                pred_label = np.argpartition(pred_label, -self.top_k)
                label = labels.cpu().numpy().astype(np.int32)
                assert (len(label) == len(pred_label))
                num_samples = pred_label.shape[0]
                num_dims = len(pred_label.shape)
                if num_dims == 1:
                    num_correct = (pred_label.flat == label.flat).sum()
                    self.sum_metric += num_correct
                    self.global_sum_metric += num_correct
                elif num_dims == 2:
                    num_classes = pred_label.shape[1]
                    top_k = min(num_classes, self.top_k)
                    for j in range(top_k):
                        num_correct = (pred_label[:, num_classes - 1 - j].flat == label.flat).sum()
                        self.sum_metric += num_correct
                        self.global_sum_metric += num_correct
                self.num_inst += num_samples
                self.global_num_inst += num_samples


class Top1Error(Accuracy):
    
    Computes top-1 error (inverted accuracy classification score).

Instances