Could you please elaborate the following line? why not use the y coming from batch?
nooralahzadeh opened this issue · comments
I think I figured it out. You consider the set of classes whose contribute in the n-shot task. However the 'y' coming from batch has all the classes. am I right?
Yes thats right. The y coming from the batch is the underlying class from the whole dataset (>1000 classes for Omniglot, 80 for miniImageNet). We don't want want to use the true class just the label of each sample within that particular n-shot task. Nicely figured out!
Yes thats right. The y coming from the batch is the underlying class from the whole dataset (>1000 classes for Omniglot, 80 for miniImageNet). We don't want want to use the true class just the label of each sample within that particular n-shot task. Nicely figured out!
But the predict of the model('logits') seems would be optimized like:
[[1, 0, 0, 0, 0], [0, 1, 0, 0, 0], [0, 0, 1, 0, 0], [0, 0, 0, 1, 0], [0, 0, 0, 0, 1]] because of the fixed label [0, 1, 2, 3, 4]. I'm so confused about that. Could you please explain this? @nooralahzadeh @oscarknagg , thanks a lot!!