Compute the accuracy by one-hot prediction and one-hot label

vision
1

Now there is:

  • prediction_one_hot:
tensor([[0, 1],
        [0, 1],
        [1, 0],
        [0, 1],
        [0, 1],
        [0, 1],
        [0, 1],
        [1, 0]], dtype=torch.uint8)
  • label_one_hot
tensor([[0, 1],
        [1, 0],
        [1, 0],
        [0, 1],
        [0, 1],
        [0, 1],
        [0, 1],
        [1, 0]])

Then how can I compute the accuracy?
This problem should not be difficult, but I have been thinking for a long time and can’t think of it. I thought there maybe a function for tensor in PyTorch can work, but I did not find that here.

2

Scikitlearn has a good implementation but it is for numpy.
Anyway you can code it yourself. The starting point :slight_smile:

def classification_metric(pred_labels, true_labels):
    pred_labels = torch.ByteTensor(pred_labels)
    true_labels = torch.ByteTensor(true_labels)

    assert 1 >= pred_labels.all() >= 0
    assert 1 >= true_labels.all() >= 0

    # True Positive (TP): we predict a label of 1 (positive), and the true label is 1.
    TP = torch.sum((pred_labels == 1) & ((true_labels == 1)))

    # True Negative (TN): we predict a label of 0 (negative), and the true label is 0.
    TN = torch.sum((pred_labels == 0) & (true_labels == 0))

    # False Positive (FP): we predict a label of 1 (positive), but the true label is 0.
    FP = torch.sum((pred_labels == 1) & (true_labels == 0))

    # False Negative (FN): we predict a label of 0 (negative), but the true label is 1.
    FN = torch.sum((pred_labels == 0) & (true_labels == 1))
    return (TP, TN, FP, FN)
1 Like
3

Thanks for your help.:upside_down_face::slightly_smiling_face: