Taking subsets of a pytorch dataset

torch.utils.data.Subset is easier, supports shuffle, and doesn’t require writing your own sampler:

import torchvision
import torch

trainset = torchvision.datasets.CIFAR10(root="./data", train=True,
                                        download=True, transform=None)

evens = list(range(0, len(trainset), 2))
odds = list(range(1, len(trainset), 2))
trainset_1 = torch.utils.data.Subset(trainset, evens)
trainset_2 = torch.utils.data.Subset(trainset, odds)

trainloader_1 = torch.utils.data.DataLoader(trainset_1, batch_size=4,
                                            shuffle=True, num_workers=2)
trainloader_2 = torch.utils.data.DataLoader(trainset_2, batch_size=4,
                                            shuffle=True, num_workers=2)