argwhere and nonzero in PyTorch

Buy Me a Coffee☕

*My post explains where() and count_nonzero().

argwhere() can get the 2D tensor of the zero or more indices of non-zero elements from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

• argwhere() can be used with torch or a tensor.
• The 1st argument(input) with torch or using a tensor(Required-Type:tensor of int, float, complex or bool).

import torch

my_tensor = torch.tensor(5)

torch.argwhere(input=my_tensor)
my_tensor.argwhere()
# tensor([], size=(1, 0), dtype=torch.int64)

my_tensor = torch.tensor([5, 0, 4, 0, 3, 1])

torch.argwhere(input=my_tensor)
# tensor([[0], [2], [4], [5]])

my_tensor = torch.tensor([5., 0., 4., 0., 3., 1])

torch.argwhere(input=my_tensor)
# tensor([[0], [2], [4], [5]])

my_tensor = torch.tensor([5.+0.j, 0.+0.j, 4.+0.j, 0.+0.j, 3.+0.j, 1.+0.j])

torch.argwhere(input=my_tensor)
# tensor([[0], [2], [4], [5]])

my_tensor = torch.tensor([True, False, True, False, True, False])

torch.argwhere(input=my_tensor)
# tensor([[0], [2], [4]])

my_tensor = torch.tensor([[5, 0, 4],
                          [0, 3, 1]])
torch.argwhere(input=my_tensor)
# tensor([[0, 0], [0, 2], [1, 1], [1, 2]])

my_tensor = torch.tensor([[[5, 0, 4], [0, 3, 1]],
                          [[0, 7, 0], [0, 6, 8]]])
torch.argwhere(input=my_tensor)
# tensor([[0, 0, 0],
#         [0, 0, 2],
#         [0, 1, 1],
#         [0, 1, 2],
#         [1, 0, 1],
#         [1, 1, 1],
#         [1, 1, 2]])

nonzero() can get the 2D tensor of the zero or more indices of non-zero elements or the one or more 1D tensors of the zero or more indices of non-zero elements from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

• nonzero() can be used with torch or a tensor.
• The 1st argument(input) with torch or using a tensor(Required-Type:tensor of int, float, complex or bool).
• There is as_tuple argument with torch or a tensor(Optional-Default:False-Type:bool): *Memos:
  • If as_tuple is True, the tuple of zero or more 1D tensors is returned.
• There is out argument with torch(Optional-Default:None-Type:tensor): *Memos:
  • out= must be used.
  • My post explains out argument.

import torch

my_tensor = torch.tensor(5)

torch.nonzero(input=my_tensor)
my_tensor.nonzero()
# tensor([], size=(1, 0), dtype=torch.int64)

torch.nonzero(input=my_tensor, as_tuple=True)
# (tensor([0]),)

my_tensor = torch.tensor([5, 0, 4, 0, 3, 1])

torch.nonzero(input=my_tensor)
my_tensor.nonzero()
# tensor([[0], [2], [4], [5]])

torch.nonzero(input=my_tensor, as_tuple=True)
# (tensor([0, 2, 4, 5]),)

my_tensor = torch.tensor([5., 0., 4., 0., 3., 1])

torch.nonzero(input=my_tensor)
# tensor([[0], [2], [4], [5]])

my_tensor = torch.tensor([5.+0.j, 0.+0.j, 4.+0.j, 0.+0.j, 3.+0.j, 1.+0.j])

torch.nonzero(input=my_tensor)
# tensor([[0], [2], [4], [5]])

my_tensor = torch.tensor([True, False, True, False, True, False])

torch.nonzero(input=my_tensor)
# tensor([[0], [2], [4]])

my_tensor = torch.tensor([[5, 0, 4],
                          [0, 3, 1]])
torch.nonzero(input=my_tensor)
# tensor([[0, 0], [0, 2], [1, 1], [1, 2]])

torch.nonzero(input=my_tensor, as_tuple=True)
# (tensor([0, 0, 1, 1]),
#  tensor([0, 2, 1, 2]))

my_tensor = torch.tensor([[[5, 0, 4], [0, 3, 1]],
                          [[0, 7, 0], [0, 6, 8]]])
torch.nonzero(input=my_tensor)
# tensor([[0, 0, 0],
#         [0, 0, 2],
#         [0, 1, 1],
#         [0, 1, 2],
#         [1, 0, 1],
#         [1, 1, 1],
#         [1, 1, 2]])

torch.nonzero(input=my_tensor, as_tuple=True)
# (tensor([0, 0, 0, 0, 1, 1, 1]),
#  tensor([0, 0, 1, 1, 0, 1, 1]),
#  tensor([0, 2, 1, 2, 1, 1, 2]))