How to do: [0,0,1,2,2,2,3,3] -> [2,1,3,2]?

1

Sorry for the confusing title. Is there a neat way to do this: given an 1-d tensor, its value starts with 0, after some zeros it changes to one, then after some ones change to two… and I want to get a tensor indicating how many zeros, ones, twos… Assume that tensor[i] <= tensor[j] iff i <= j.

E.g. [0,0,1,2,2,2,3,3] → [2,1,3,2], there are two zeros, one one, three twos, and two threes, so the target tensor is [2,1,3,2].

2 
a = torch.tensor([0,0,1,2,2,2,3,3])
(a == torch.arange(a.max()+1)[:, None]).sum(dim=1)

Found one way.