The topk operation in Pytorch "Returns the k largest elements of the given input tensor along a given dimension." (from here). But does it have an opposite that returns the smallest k elements?

top k

x = torch.arange(1., 6.)
print(x)
>>> tensor([ 1.,  2.,  3.,  4.,  5.])
torch.topk(x, 3)
>>> torch.return_types.topk(values=tensor([5., 4., 3.]), indices=tensor([4, 3, 2]))

min k –> does this exist?

x = torch.arange(1., 6.)
print(x)
>>> tensor([ 1.,  2.,  3.,  4.,  5.])
torch.mink(x, 3) # doesn't run
>>> torch.return_types.mink(values=tensor([1., 2., 3.]), indices=tensor([0, 1, 2]))

I know that a way around it is to multiply the tensor by -1:

torch.topk(-x, 3)

But wondering if this mink operation already exist.

>Solution :

Check the documention page https://pytorch.org/docs/stable/generated/torch.topk.html

There is a keyword argument to get the smallest values (and indices)

largest (bool, optional) – controls whether to return largest or smallest elements

so mink is just functools.partial(torch.topk, largest=False)