DEV Community

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

Posted on • Edited on

sort, argsort and msort in PyTorch

#python #pytorch #sort #msort

Buy Me a Coffee☕

sort() can get two of the 0D or more D tensors of zero or more sorted elements and their indices in ascending(Default) or descending order from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

  • sort() 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 0D tensor of a complex number can be used.
  • The 2nd argument with torch or the 1st argument with a tensor is dim(Optional-Defualt:-1-Type:int).
  • The 3rd argument with torch or the 2nd argument with a tensor is descending(Optional-Default:False-Type:bool). *False is ascending order and True is descending order.
  • The 4th argument with torch or the 3rd argument with a tensor is stable(Optional-Default:False-Type:bool). *Memos:
    • True can definitely sort the multiple same values while False cannot but even False basically can sort the multiple same values.
    • You must use stable=. *Setting stable, dim and descending also needs dim= and descending= respectively.
  • There is out argument with torch(Optional-Default:None-Type:tuple(tensor, tensor) or list(tensor, tensor)): *Memos:
    • out= must be used.
    • My post explains out argument. 
import torch

my_tensor = torch.tensor([7, 1, -5, 7, 9, -3, 0, -3])

torch.sort(my_tensor)
my_tensor.sort()
torch.sort(input=my_tensor, dim=0, descending=False, stable=False)
torch.sort(input=my_tensor, dim=-1, descending=False, stable=False)
torch.sort(input=my_tensor, dim=0, descending=False, stable=True)
torch.sort(input=my_tensor, dim=-1, descending=False, stable=True)
# torch.return_types.sort(
# values=tensor([-5, -3, -3, 0, 1, 7, 7, 9]),
# indices=tensor([2, 5, 7, 6, 1, 0, 3, 4]))

torch.sort(input=my_tensor, dim=0, descending=True, stable=False)
torch.sort(input=my_tensor, dim=-1, descending=True, stable=False)
# torch.return_types.sort(
# values=tensor([9, 7, 7, 1, 0, -3, -3, -5]),
# indices=tensor([4, 0, 3, 1, 6, 5, 7, 2]))

my_tensor = torch.tensor([[7, 1, -5, 7], [9, -3, 0, -3]])

torch.sort(input=my_tensor)
torch.sort(input=my_tensor, dim=-1, descending=False, stable=False)
torch.sort(input=my_tensor, dim=1, descending=False, stable=False)
torch.sort(input=my_tensor, dim=-1, descending=False, stable=True)
torch.sort(input=my_tensor, dim=1, descending=False, stable=True)
# torch.return_types.sort(
# values=tensor([[-5, 1, 7, 7], [-3, -3, 0, 9]]),
# indices=tensor([[2, 1, 0, 3], [1, 3, 2, 0]]))

torch.sort(input=my_tensor, dim=0, descending=False, stable=False)
torch.sort(input=my_tensor, dim=-2, descending=False, stable=False)
torch.sort(input=my_tensor, dim=0, descending=False, stable=True)
torch.sort(input=my_tensor, dim=-2, descending=False, stable=True)
# torch.return_types.sort(
# values=tensor([[7, -3, -5, -3], [9, 1, 0, 7]]),
# indices=tensor([[0, 1, 0, 1], [1, 0, 1, 0]]))

my_tensor = torch.tensor([[7., 1, -5., 7.], [9., -3., 0, -3.]])

torch.sort(input=my_tensor)
# torch.return_types.sort(
# values=tensor([[-5., 1., 7., 7.], [-3., -3., 0., 9.]]),
# indices=tensor([[2, 1, 0, 3], [1, 3, 2, 0]]))

my_tensor = torch.tensor([[True, False, True, False],
                          [False, True, False, True]])
torch.sort(input=my_tensor)
# torch.return_types.sort(
# values=tensor([[False, False, True,  True],
#                [False, False, True,  True]]),
# indices=tensor([[1, 3, 0, 2],
#                 [0, 2, 1, 3]]))

my_tensor = torch.tensor(7.+0.j)

torch.sort(input=my_tensor)
# torch.return_types.sort(
# values=tensor(7.+0.j),
# indices=tensor(0))
Enter fullscreen mode Exit fullscreen mode

argsort() can get the 0D or more D tensor of zero or more sorted elements' indices in ascending(Default) or descending order from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

  • argsort() 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 0D tensor of a complex number can be used.
  • The 2nd argument with torch or the 1st argument with a tensor is dim(Optional-Defualt:-1Type:int).
  • The 3rd argument with torch or the 2nd argument with a tensor is descending(Optional-Default:False-Type:bool). *False is ascending order and True is descending order.
  • The 4th argument with torch or the 3rd argument with a tensor is stable(Optional-Default:False-Type:bool). *Memos:
    • True can definitely sort the multiple same values while False cannot but even False can basically sort the multiple same values.
    • You must use stable=. *When setting stable, dim and descending also need dim= and descending= respectively. 
import torch

my_tensor = torch.tensor([7, 1, -5, 7, 9, -3, 0, -3])

torch.argsort(input=my_tensor)
my_tensor.argsort()
torch.argsort(input=my_tensor, dim=0, descending=False, stable=False)
torch.argsort(input=my_tensor, dim=-1, descending=False, stable=False)
torch.argsort(input=my_tensor, dim=0, descending=False, stable=True)
torch.argsort(input=my_tensor, dim=-1, descending=False, stable=True)
# tensor([2, 5, 7, 6, 1, 0, 3, 4])

torch.argsort(input=my_tensor, dim=0, descending=True, stable=False)
torch.argsort(input=my_tensor, dim=-1, descending=True, stable=False)
# tensor([4, 0, 3, 1, 6, 5, 7, 2])

my_tensor = torch.tensor([[7, 1, -5, 7], [9, -3, 0, -3]])

torch.argsort(input=my_tensor)
torch.argsort(input=my_tensor, dim=-1, descending=False, stable=False)
torch.argsort(input=my_tensor, dim=1, descending=False, stable=False)
torch.argsort(input=my_tensor, dim=-1, descending=False, stable=True)
torch.argsort(input=my_tensor, dim=1, descending=False, stable=True)
# tensor([[2, 1, 0, 3], [1, 3, 2, 0]])

torch.argsort(input=my_tensor, dim=0, descending=False, stable=False)
torch.argsort(input=my_tensor, dim=-2, descending=False, stable=False)
torch.argsort(input=my_tensor, dim=0, descending=False, stable=True)
torch.argsort(input=my_tensor, dim=-2, descending=False, stable=True)
# tensor([[0, 1, 0, 1], [1, 0, 1, 0]])

my_tensor = torch.tensor([[7., 1., -5., 7.], [9., -3., 0., -3.]])

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

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

my_tensor = torch.tensor(7.+0.j)

torch.argsort(input=my_tensor)
# tensor(0)
Enter fullscreen mode Exit fullscreen mode

msort() can get the 0D or more D tensor of zero or more sorted elements in ascending order from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

  • msort() 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 0D tensor of a complex number can be used.
  • msort() doesn't have dim, descending and stable argument.
  • msort() is equivalent to sort()[0] with dim=0.
  • There is out argument with torch(Optional-Default:None-Type:tuple(tensor, tensor) or list(tensor, tensor)): *Memos:
    • out= must be used.
    • My post explains out argument. 
import torch

my_tensor = torch.tensor([7, 1, -5, 7, 9, -3, 0, -3])

torch.msort(input=my_tensor)
my_tensor.msort()
# tensor([-5, -3, -3, 0, 1, 7, 7, 9])

my_tensor = torch.tensor([[7, 1, -5, 7], [9, -3, 0, -3]])

torch.msort(input=my_tensor)
# tensor([[7, -3, -5, -3], [9, 1, 0, 7]])

my_tensor = torch.tensor([[7., 1., -5., 7.], [9., -3., 0., -3.]])

torch.msort(input=my_tensor)
# tensor([[7., -3., -5., -3.], [9., 1., 0., 7.]])

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

my_tensor = torch.tensor(7.+5.j)

torch.msort(input=my_tensor)
# tensor(7.+5.j)
Enter fullscreen mode Exit fullscreen mode

Top comments (0)