DEV Community

Super Kai (Kazuya Ito)
Super Kai (Kazuya Ito)

Posted on • Edited on

unique and unique_consecutive in PyTorch

Buy Me a Coffee

*[Warning]-unique() and unique_consecutive() are really tricky.

unique() can get the one or more 1D or more D tensors of zero or more unique elements, their indices and their counts as shown below:

*Memos:

  • unique() can be used with torch or a tensor.
  • The 1st argument(input) with torch or using a tensor(Required-Type:tensor of int, float or bool).
  • The 2nd argument with torch or the 1st argument with a tensor is sorted(Optional-Default:True-Type:bool). *Currently, setting False is also sorted.
  • The 3rd argument with torch or the 2nd argument with a tensor is return_inverse(Optional-Default:False-Type:bool). *It returns the indices for where elements in the original input ended up in the returned unique list.
  • The 4th argument with torch or the 3rd argument with a tensor is return_counts(Optional-Default:False-Type:bool). *It returns the counts for each unique element.
  • The 5th argument with torch or the 4th argument with a tensor is dim(Optional-Type:int).
import torch

my_tensor = torch.tensor([[[2, 2, 0], [0, 1, 1]],
                          [[1, 3, 0], [0, 0, 2]]])
torch.unique(input=my_tensor)
my_tensor.unique()
# tensor([0, 1, 2, 3])

torch.unique(input=my_tensor,
             sorted=False,
             return_inverse=True,
             return_counts=True)
# (tensor([0, 1, 2, 3]),
#  tensor([[[2, 2, 0], [0, 1, 1]],
#          [[1, 3, 0], [0, 0, 2]]]),
#  tensor([5, 3, 3, 1]))

torch.unique(input=my_tensor,
             sorted=False,
             return_inverse=True,
             return_counts=True,
             dim=0)
torch.unique(input=my_tensor,
             sorted=False,
             return_inverse=True,
             return_counts=True,
             dim=-3)
# (tensor([[[1, 3, 0], [0, 0, 2]],
#           [[2, 2, 0], [0, 1, 1]]]),
#  tensor([1, 0]),
#  tensor([1, 1]))

torch.unique(input=my_tensor,
             sorted=False,
             return_inverse=True,
             return_counts=True,
             dim=1)
torch.unique(input=my_tensor,
             sorted=False,
             return_inverse=True,
             return_counts=True,
             dim=-2)
# (tensor([[[0, 1, 1], [2, 2, 0]],
#          [[0, 0, 2], [1, 3, 0]]]),
#  tensor([1, 0]),
#  tensor([1, 1]))

torch.unique(input=my_tensor,
             sorted=False,
             return_inverse=True,
             return_counts=True,
             dim=2)
torch.unique(input=my_tensor,
             sorted=False,
             return_inverse=True,
             return_counts=True,
             dim=-1)
# (tensor([[[0, 2, 2], [1, 0, 1]],
#          [[0, 1, 3], [2, 0, 0]]]),
#  tensor([1, 2, 0]),
#  tensor([1, 1, 1]))

my_tensor = torch.tensor([[[2., 2., 0.], [0., 1., 1.]],
                          [[1., 3., 0.], [0., 0., 2.]]])
torch.unique(input=my_tensor)
# tensor([0., 1., 2., 3.])

my_tensor = torch.tensor([[[True, False, True], [True, False, True]],
                          [[False, True, False], [False, True, False]]])
torch.unique(input=my_tensor)
# tensor([False, True])
Enter fullscreen mode Exit fullscreen mode

unique_consecutive() can get the zero or more unique elements of a 0D or more D tensor by consecutiveness as shown below:

*Memos:

  • unique_consecutive() can be used with torch or a tensor.
  • The 1st argument(input) with torch or using a tensor(Required-Type:tensor of int, float or bool).
  • The 2nd argument with torch or the 1st argument with a tensor is return_inverse(Optional-Default:False-Type:bool). *It returns the indices for where elements in the original input ended up in the returned unique list.
  • The 3rd argument with torch or the 2nd argument with a tensor is return_counts(Optional-Default:False-Type:bool). *It returns the counts for each unique element.
  • The 4th argument with torch or the 3rd argument with a tensor is dim(Optional-Type:int).
  • unique_consecutive() doesn't have sorted argument.
import torch

my_tensor = torch.tensor([[[2, 2, 0], [0, 1, 1]],
                          [[1, 3, 0], [0, 0, 2]]])
torch.unique_consecutive(input=my_tensor)
my_tensor.unique_consecutive()
# tensor([2, 0, 1, 3, 0, 2])

torch.unique_consecutive(input=my_tensor,
                         return_inverse=True,
                         return_counts=True)
# (tensor([2, 0, 1, 3, 0, 2]),
#  tensor([[[0, 0, 1], [1, 2, 2]],
#          [[2, 3, 4], [4, 4, 5]]]),
#  tensor([2, 2, 3, 1, 3, 1]))

torch.unique_consecutive(input=my_tensor,
                         return_inverse=True,
                         return_counts=True,
                         dim=0)
torch.unique_consecutive(input=my_tensor,
                         return_inverse=True,
                         return_counts=True,
                         dim=-3)
# (tensor([[[2, 2, 0], [0, 1, 1]],
#          [[1, 3, 0], [0, 0, 2]]]),
#  tensor([0, 1]),
#  tensor([1, 1]))

torch.unique_consecutive(input=my_tensor,
                         return_inverse=True,
                         return_counts=True,
                         dim=1)
torch.unique_consecutive(my_tensor,
                         return_inverse=True,
                         return_counts=True,
                         dim=-2)
# (tensor([[[2, 2, 0], [0, 1, 1]],
#          [[1, 3, 0], [0, 0, 2]]]),
#  tensor([0, 1]),
#  tensor([1, 1]))

torch.unique_consecutive(input=my_tensor,
                         return_inverse=True,
                         return_counts=True,
                         dim=2)
torch.unique_consecutive(input=my_tensor,
                         return_inverse=True,
                         return_counts=True,
                         dim=-1)
# (tensor([[[2, 2, 0], [0, 1, 1]],
#         [[1, 3, 0], [0, 0, 2]]]),
#  tensor([0, 1, 2]),
#  tensor([1, 1, 1]))

my_tensor = torch.tensor([[[2., 2., 0.], [0., 1., 1.]],
                          [[1., 3., 0.], [0., 0., 2.]]])
torch.unique_consecutive(input=my_tensor)
# tensor([2., 0., 1., 3., 0., 2.])

my_tensor = torch.tensor([[[True, False, True], [True, False, True]],
                          [[False, True, False], [False, True, False]]])
torch.unique_consecutive(input=my_tensor)
# tensor([True, False, True, False, True, False, True, False, True, False])
Enter fullscreen mode Exit fullscreen mode

Speedy emails, satisfied customers

Postmark Image

Are delayed transactional emails costing you user satisfaction? Postmark delivers your emails almost instantly, keeping your customers happy and connected.

Sign up

Top comments (0)

Image of Docusign

🛠️ Bring your solution into Docusign. Reach over 1.6M customers.

Docusign is now extensible. Overcome challenges with disconnected products and inaccessible data by bringing your solutions into Docusign and publishing to 1.6M customers in the App Center.

Learn more