DEV Community

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

Posted on • Edited on

nan_to_num in PyTorch

#python #pytorch #nantonum #function

Buy Me a Coffee☕

*My post explains torch.nan and torch.inf.

nan_to_num() can get the 0D or more D tensor of zero or more elements, replacing zero or more NaNs(Not a Numbers), positive infinities and negative infinities with zero or more zeros, the greatest finities and the least finities respectively(Default) or specified values from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

  • nan_to_num() 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 nan(Optional-Default:Zero-Type:int, float or bool).
  • The 3rd argument with torch or the 2nd argument with a tensor is posinf(Optional-Default:The greatest finite-Type:int, float or bool).
  • The 4th argument with torch or the 2nd argument with a tensor is neginf(Optional-Default:The lowest finite-Type:int, float or bool).
  • 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([-torch.inf, 7., -5., torch.inf, 8., torch.nan, torch.inf, torch.nan]) torch.nan_to_num(input=my_tensor) my_tensor.nan_to_num() # tensor([-3.4028e+38, 7.0000e+00, -5.0000e+00, 3.4028e+38, # 8.0000e+00, 0.0000e+00, 3.4028e+3, 0.0000e+00])  torch.nan_to_num(input=my_tensor, nan=2., posinf=-6., neginf=9.) # tensor([9., 7., -5., -6., 8., 2., -6., 2.])  my_tensor = torch.tensor([[-torch.inf, 7., -5., torch.inf], [8., torch.nan, torch.inf, torch.nan]]) torch.nan_to_num(input=my_tensor, nan=2., posinf=-6., neginf=9.) # tensor([[9., 7., -5., -6.], # [8., 2., -6., 2.]])  my_tensor = torch.tensor([[[-torch.inf, 7.], [-5., torch.inf]], [[8., torch.nan], [torch.inf, torch.nan]]]) torch.nan_to_num(input=my_tensor, nan=2., posinf=-6., neginf=9.) # tensor([[[9., 7.], [-5., -6.]], # [[8., 2.], [-6., 2.]]])  my_tensor = torch.tensor([complex(-torch.inf, torch.inf), 7.+0.j, -5.+0.j, complex(torch.inf, -torch.inf), 8.+0.j, complex(torch.nan, torch.nan), complex(torch.inf), complex(torch.nan)]) torch.nan_to_num(input=my_tensor) # tensor([-3.4028e+38+3.4028e+38j, 7.0000e+00+0.0000e+00j, # -5.0000e+00+0.0000e+00j, 3.4028e+38-3.4028e+38j, # 8.0000e+00+0.0000e+00j, 0.0000e+00+0.0000e+00j, # 3.4028e+38+0.0000e+00j, 0.0000e+00+0.0000e+00j])  torch.nan_to_num(input=my_tensor, nan=2., posinf=-6., neginf=9.) # tensor([9.-6.j, 7.+0.j, # -5.+0.j, -6.+9.j, # 8.+0.j, 2.+2.j, # -6.+0.j, 2.+0.j])  my_tensor = torch.tensor([[complex(-torch.inf, torch.inf), 7.+0.j, -5.+0.j, complex(torch.inf, -torch.inf)], [8.+0.j, complex(torch.nan, torch.nan), complex(torch.inf), complex(torch.nan)]]) torch.nan_to_num(input=my_tensor, nan=2., posinf=-6., neginf=9.) # tensor([[9.-6.j, 7.+0.j, # -5.+0.j, -6.+9.j], # [8.+0.j, 2.+2.j, # -6.+0.j, 2.+0.j]])  my_tensor = torch.tensor([[[complex(-torch.inf, torch.inf), 7.+0.j], [-5.+0.j, complex(torch.inf, -torch.inf)]], [[8.+0.j, complex(torch.nan, torch.nan)], [complex(torch.inf), complex(torch.nan)]]]) torch.nan_to_num(input=my_tensor, nan=2., posinf=-6., neginf=9.) # tensor([[[9.-6.j, 7.+0.j], # [-5.+0.j, -6.+9.j]], # [[8.+0.j, 2.+2.j], # [-6.+0.j, 2.+0.j]]])  my_tensor = torch.tensor([[[0, 1], [2, 3]], [[4, 5], [6, 7]]]) torch.nan_to_num(input=my_tensor, nan=2, posinf=-6, neginf=9) # tensor([[[0, 1], [2, 3]], [[4, 5], [6, 7]]])  my_tensor = torch.tensor([[[True, False], [True, False]], [[False, True], [False, True]]]) torch.nan_to_num(input=my_tensor, nan=True, posinf=False, neginf=True) # tensor([[[True, False], [True, False]], # [[False, True], [False, True]]])
Enter fullscreen mode Exit fullscreen mode

Top comments (0)