DEV Community

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

Posted on • Edited on

sum and nansum in PyTorch

#python #pytorch #sum #nansum

Buy Me a Coffee☕

*Memos:

sum() can get the 0D or more D tensor of zero or more sum's elements, normally treating one or more NaNs(Not a Numbers) from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

  • sum() 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).
  • The 2nd argument with torch or the 1st argument with a tensor is dim(Optional-Type:int, tuple of int or list of int).
  • The 3rd argument with torch or the 2nd argument with a tensor is keepdim(Optional-Default:False-Type:bool): *Memos:
    • keepdim= must be used with dim=.
    • My post explains keepdim argument.
  • There is dtype argument with torch(Optional-Default:None-Type:dtype): *Memos:
    • If it's None, it's inferred from input or a tensor.
    • dtype= must be used.
    • My post explains dtype argument.
  • Normally, the arithmetic operation with a NaN results in a NaN.
  • The empty 1D or more D input tensor or tensor without dim or with the deepest dim gets a zero. 
import torch my_tensor = torch.tensor([1, 2, 3, 4]) torch.sum(input=my_tensor) my_tensor.sum() torch.sum(input=my_tensor, dim=0) torch.sum(input=my_tensor, dim=-1) torch.sum(input=my_tensor, dim=(0,)) torch.sum(input=my_tensor, dim=(-1,)) # tensor(10)  my_tensor = torch.tensor([1, 2, torch.nan, 4]) torch.sum(input=my_tensor) # tensor(nan)  my_tensor = torch.tensor([[1, 2, 3, 4], [5, 6, 7, 8]]) torch.sum(input=my_tensor) torch.sum(input=my_tensor, dim=(0, 1)) torch.sum(input=my_tensor, dim=(0, -1)) torch.sum(input=my_tensor, dim=(1, 0)) torch.sum(input=my_tensor, dim=(1, -2)) torch.sum(input=my_tensor, dim=(-1, 0)) torch.sum(input=my_tensor, dim=(-1, -2)) torch.sum(input=my_tensor, dim=(-2, 1)) torch.sum(input=my_tensor, dim=(-2, -1)) # tensor(36)  torch.sum(input=my_tensor, dim=0) torch.sum(input=my_tensor, dim=-2) torch.sum(input=my_tensor, dim=(0,)) torch.sum(input=my_tensor, dim=(-2,)) # tensor([6, 8, 10, 12])  torch.sum(input=my_tensor, dim=1) torch.sum(input=my_tensor, dim=-1) torch.sum(input=my_tensor, dim=(1,)) torch.sum(input=my_tensor, dim=(-1,)) # tensor([10, 26])  my_tensor = torch.tensor([[1, 2, torch.nan, 4], [torch.nan, 6, 7, 8]]) torch.sum(input=my_tensor) # tensor(nan)  torch.sum(input=my_tensor, dim=0) # tensor([nan, 8., nan, 12.])  torch.sum(input=my_tensor, dim=1) # tensor([nan, nan])  my_tensor = torch.tensor([[1., 2., 3., 4.], [5., 6., 7., 8.]]) torch.sum(input=my_tensor) # tensor(36.)  my_tensor = torch.tensor([[1, 2, torch.nan, 4], [torch.nan, 6, 7, 8]]) torch.sum(input=my_tensor) # tensor(nan)  my_tensor = torch.tensor([[1.+0.j, 2.+0.j, 3.+0.j, 4.+0.j], [5.+0.j, 6.+0.j, 7.+0.j, 8.+0.j]]) torch.sum(input=my_tensor) # tensor(36.+0.j)  my_tensor = torch.tensor([[1.+0.j, 2.+0.j, torch.nan, 4.+0.j], [torch.nan, 6.+0.j, 7.+0.j, 8.+0.j]]) torch.sum(input=my_tensor) # tensor(nan+0.j)  my_tensor = torch.tensor([[True, False, True, False], [False, True, False, True]]) torch.sum(input=my_tensor) # tensor(4)  my_tensor = torch.tensor([]) torch.sum(input=my_tensor) # tensor(0.)
Enter fullscreen mode Exit fullscreen mode

nansum() can get the 0D or more D tensor of zero or more sum's elements, treating nan as zero from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

  • nansum() 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 dim(Optional-Type:int, tuple of int or list of int).
  • The 3rd argument with torch or the 2nd argument with a tensor is keepdim(Optional-Default:False-Type:bool): *Memos:
    • keepdim= must be used with dim=.
    • My post explains keepdim argument.
  • There is dtype argument with torch(Optional-Default:None-Type:dtype): *Memos:
    • If it's None, it's inferred from input or a tensor.
    • dtype= must be used.
    • My post explains dtype argument.
  • Normally, the arithmetic operation with a NaN results in a NaN.
  • The empty 1D or more D input tensor or tensor without dim or with the deepest dim gets a zero. 
import torch my_tensor = torch.tensor([1., 2., torch.nan, 4.]) torch.nansum(input=my_tensor) my_tensor.nansum() torch.nansum(input=my_tensor, dim=0) torch.nansum(input=my_tensor, dim=-1) torch.nansum(input=my_tensor, dim=(0,)) torch.nansum(input=my_tensor, dim=(-1,)) # tensor(7.)  my_tensor = torch.tensor([[1., 2., torch.nan, 4.], [torch.nan, 6., 7., 8.]]) torch.nansum(input=my_tensor) torch.nansum(input=my_tensor, dim=(0, 1)) torch.nansum(input=my_tensor, dim=(0, -1)) torch.nansum(input=my_tensor, dim=(1, 0)) torch.nansum(input=my_tensor, dim=(1, -2)) torch.nansum(input=my_tensor, dim=(-1, 0)) torch.nansum(input=my_tensor, dim=(-1, -2)) torch.nansum(input=my_tensor, dim=(-2, 1)) torch.nansum(input=my_tensor, dim=(-2, -1)) # tensor(28.)  torch.nansum(input=my_tensor, dim=0) torch.nansum(input=my_tensor, dim=-2) torch.nansum(input=my_tensor, dim=(0,)) torch.nansum(input=my_tensor, dim=(-2,)) # tensor([1., 8., 7., 12.])  torch.nansum(input=my_tensor, dim=1) torch.nansum(input=my_tensor, dim=-1) torch.nansum(input=my_tensor, dim=(1,)) torch.nansum(input=my_tensor, dim=(-1,)) # tensor([7., 21.])  my_tensor = torch.tensor([[1, 2, 3, 4], [5, 6, 7, 8]]) torch.nansum(input=my_tensor) # tensor(36)  my_tensor = torch.tensor([[True, False, True, False], [False, True, False, True]]) torch.nansum(input=my_tensor) # tensor(4)  my_tensor = torch.tensor([]) torch.nansum(input=my_tensor) # tensor(0.)
Enter fullscreen mode Exit fullscreen mode

Top comments (0)