Selecting support and query sets randomly
Closed this issue · 1 comments
Harut0726 commented
Thank you for sharing your code.
If I understood correctly, you don't sample support and query sets randomly unlike the original paper of Protonet. Can this be the reason of overfitting on custom dataset?
data, _ = [_.cuda() for _ in batch]
p = args.shot * args.train_way
data_shot, data_query = data[:p], data[p:]
Harut0726 commented
Sorry, actually now I just noticed that you do select them randomly.
pos = torch.randperm(len(l))[:self.n_per]