As far as I see, you have been using torch.randperm in support-query sets selection. Since torch.randperm returns random integer numbers, should we be sure that the distribution of each class is uniform both in train and test sections? For example, is this possible to miss one of the train classes during training or use too many times one of the test classes during the test section?
Or, is this strategy widely used in the few shot data learning algorithms?
Hi,
As far as I see, you have been using torch.randperm in support-query sets selection. Since torch.randperm returns random integer numbers, should we be sure that the distribution of each class is uniform both in train and test sections? For example, is this possible to miss one of the train classes during training or use too many times one of the test classes during the test section?
Or, is this strategy widely used in the few shot data learning algorithms?
Thank you