_unique_dim does not support zero size tensor #18408
Description
Unique without dim works well
>>> a = torch.empty(5, 0)
>>> a
tensor([], size=(5, 0))
>>> a.unique(return_inverse=True)
(tensor([]), tensor([], size=(5, 0), dtype=torch.int64))But unique with dim does not work
>>> a.unique(dim=1, return_inverse=True)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/Users/gaoxiang/anaconda3/lib/python3.7/site-packages/torch/tensor.py", line 351, in unique
dim=dim
RuntimeError: cannot reshape tensor of 0 elements into shape [0, -1]
>>> a.unique(dim=0, return_inverse=True)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/Users/gaoxiang/anaconda3/lib/python3.7/site-packages/torch/tensor.py", line 351, in unique
dim=dim
RuntimeError: cannot reshape tensor of 0 elements into shape [-1, 0]I suggest changing the behavior to:
if dim refers to a dimension with size 0, and other dimensions have size > 0, then the output and reverse_indicies should be the same empty tensor as input. If dim refers to a dimension with size > 0, or there are other 0 sized dimensions, then reject the input with an error because whether "nothing == nothing" is not defined.