accuracy function only gives zero-output and breaks for new numpy version

[pbouss]

The accuracy function in metrics.py does not work properly.

def accuracy(y_pred: Tensor, y_true: Tensor, **kwargs):
y_pred = y_pred.cpu()
outputs = np.argmax(y_pred, axis=1)
return np.mean(outputs.numpy() == y_true.detach().cpu().numpy())

This is due to outputs having less axes than y_true. In the newest numpy 1.26 this raises in error. In older numpy versions this gives an DeprecationWarning and outputs False, due to the mean it becomes 0. Therefore this functions outputs constantly zero.
Possibly it could be adapted easily.