How to use metrics with multi labels

[FrancescoSaverioZuppichini]

Hi guys,

I hope you are doing great! I was wondering how to correctly use a metric, e.g. Accuracy, with multi labels. For example

num_classes = 4

pred = torch.tensor([0, 0, 0, 0])
target = torch.tensor([0, 1, 0, 1])

acc = M.Accuracy(num_classes=num_classes)
print(acc(pred, target))

Where target is a one-hot vector and if target[i] == 1 it means that we have class i in the target. In this case, accuracy doesn't work, since it only compares the raw data inside it.

Thanks!

[SkafteNicki]

For multi-labels we assume that you are providing data as [B,L] where B is the batch size and L is the number of labels (num_classes in your code). You are therefore missing the batch dimension, which is 1 in your case. A simple unsqueeze should be enough:

pred = torch.tensor([0, 0, 0, 0]).unsqueeze(0)
target = torch.tensor([0, 1, 0, 1]).unsqueeze(0)