Torch equivalent of numpy.random.choice?

is there a reason why you prefer:

rand_choices=samples[torch.randint(len(samples),(7,))] #’'7 choices ‘’

vs

**perm = torch.randperm(tensor.size(0))**

?

Just curious