In torch.utils.data.sampler there is RandomSampler and SubsetRandomSampler but not SubsetSampler in cases you don't want random sampling from the indices. Any reasons for this? If it is just a missing feature, I can send a PR.