I am trying to separate pytorch MNIST dataset into n subsets each containing data from only 2 classes and then save the n subsets in differents files to load them later.
For example, subset 0 will contain images of class 0 and 1 (corresponding to the MNIST classes), subset 1 will contain images of class 1 and 2 and so on up to subset n.
If n is bigger than the number of classes several subsets will contain the same classes combination, ex: subset 10 can also contain images from class 0 and 1.
But every datapoint is given to only one subset, the same data point can not be in subset 0 and 10 for example.
I tried to write a code that does that using torch.utils.data.random_split and torch.utils.data.ConcatDataset but strangely in the files I save there always is the same data…
Thank you for your Help,