DEV Community

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

Posted on • Edited on

min and max in PyTorch

#python #pytorch #min #max

Buy Me a Coffee☕

*Memos:

min() can get the 0D of the 1st one minimum element or two of the 0D or more D tensors of the 1st zero or more minimum elements and their indices from the one or two 0D or more D tensors of zero or more elements as shown below:

*Memos:

  • min() 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 is dim(Optional-Type:int). *Setting dim can get zero or more 1st minimum elements and their indices.
  • The 2nd argument with torch or the 1st argument is other(Optional-Type:tensor of int, float or bool): *Memos:
    • It can only be used with input.
    • This is the functionality of minimum().
  • The 3rd argument with torch or the 2nd argument is keepdim(Optional-Default:False-Type:bool): *Memos:
    • It must be used with dim and without other.
    • My post explains keepdim argument.
  • There is out argument with torch(Optional-Default:None-Type:tensor, tuple(tensor, tensor) or list(tensor, tensor)): *Memos:
    • The type of tensor must be used without dim and keepdim.
    • The type of tuple(tensor, tensor) or list(tensor, tensor) must be used with dim and without other.
    • out= must be used.
    • My post explains out argument.
  • The empty 2D or more D tensor without other tensor doesn't work if not setting dim.
  • The empty 1D tensor without other tensor doesn't work even if setting dim. 
import torch my_tensor = torch.tensor([[5, 4, 7, 7], [6, 5, 3, 5], [3, 8, 9, 3]]) torch.min(input=my_tensor) my_tensor.min() # tensor(3)  torch.min(input=my_tensor, dim=0) torch.min(input=my_tensor, dim=-2) # torch.return_types.min( # values=tensor([3, 4, 3, 3]), # indices=tensor([2, 0, 1, 2]))  torch.min(input=my_tensor, dim=1) torch.min(input=my_tensor, dim=-1) # torch.return_types.min( # values=tensor([4, 3, 3]), # indices=tensor([1, 2, 0]))  tensor1 = torch.tensor([5, 4, 7, 7]) tensor2 = torch.tensor([[6, 5, 3, 5], [3, 8, 9, 3]]) torch.min(input=tensor1, other=tensor2) # tensor([[5, 4, 3, 5], # [3, 4, 7, 3]])  tensor1 = torch.tensor([5., 4., 7., 7.]) tensor2 = torch.tensor([[6., 5., 3., 5.], [3., 8., 9., 3.]]) torch.min(input=tensor1, other=tensor2) # tensor([[5., 4., 3., 5.], # [3., 4., 7., 3.]])  tensor1 = torch.tensor([True, False, True, False]) tensor2 = torch.tensor([[True, False, True, False], [False, True, False, True]]) torch.min(input=tensor1, other=tensor2) # tensor([[True, False, True, False], # [False, False, False, False]])  my_tensor = torch.tensor([]) my_tensor = torch.tensor([[]]) my_tensor = torch.tensor([[[]]]) torch.min(input=my_tensor) # Error  my_tensor = torch.tensor([]) torch.min(input=my_tensor, dim=0) # Error  my_tensor = torch.tensor([[]]) torch.min(input=my_tensor, dim=0) # torch.return_types.min( # values=tensor([]), # indices=tensor([], dtype=torch.int64))  my_tensor = torch.tensor([[[]]]) torch.min(input=my_tensor, dim=0) # torch.return_types.min( # values=tensor([], size=(1, 0)), # indices=tensor([], size=(1, 0), dtype=torch.int64))
Enter fullscreen mode Exit fullscreen mode

max() can get the 0D of the 1st one maximum element or two of the 0D or more D tensors of the 1st zero or more maximum elements and their indices from the one or two 0D or more D tensors of zero or more elements as shown below:

  • max() 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 is dim(Optional-Type:int). *Setting dim can get zero or more 1st maximum elements and their indices.
  • The 2nd argument with torch or the 1st argument is other(Optional-Type:tensor of int, float or bool): *Memos:
    • It can only be used with input.
    • This is the functionality of maximum().
  • The 3rd argument with torch or the 2nd argument is keepdim(Optional-Default:False-Type:bool): *Memos:
    • It must be used with dim and without other.
    • My post explains keepdim argument.
  • There is out argument with torch(Optional-Default:None-Type:tensor, tuple(tensor, tensor) or list(tensor, tensor)): *Memos:
    • The type of tensor must be used without dim and keepdim.
    • The type of tuple(tensor, tensor) or list(tensor, tensor) must be used with dim and without other.
    • out= must be used.
    • My post explains out argument.
  • The empty 2D or more D tensor without other tensor doesn't work if not setting dim.
  • The empty 1D input tesnor or tensor without other tensor doesn't work even if setting dim. 
import torch my_tensor = torch.tensor([[5, 4, 7, 7], [6, 5, 3, 5], [3, 8, 9, 3]]) torch.max(input=my_tensor) my_tensor.max() # tensor(9)  torch.max(input=my_tensor, dim=0) torch.max(input=my_tensor, dim=-2) # torch.return_types.max( # values=tensor([6, 8, 9, 7]), # indices=tensor([1, 2, 2, 0]))  torch.max(input=my_tensor, dim=1) torch.max(input=my_tensor, dim=-1) # torch.return_types.max( # values=tensor([7, 6, 9]), # indices=tensor([2, 0, 2]))  tensor1 = torch.tensor([5, 4, 7, 7]) tensor2 = torch.tensor([[6, 5, 3, 5], [3, 8, 9, 3]]) torch.max(input=tensor1, other=tensor2) # tensor([[6, 5, 7, 7], # [5, 8, 9, 7]])  tensor1 = torch.tensor([5., 4., 7., 7.]) tensor2 = torch.tensor([[6., 5., 3., 5.], [3., 8., 9., 3.]]) torch.max(input=tensor1, other=tensor2) # tensor([[6., 5., 7., 7.], # [5., 8., 9., 7.]])  tensor1 = torch.tensor([True, False, True, False]) tensor2 = torch.tensor([[True, False, True, False], [False, True, False, True]]) torch.max(input=tensor1, other=tensor2) # tensor([[True, False, True, False], # [True, True, True, True]])  my_tensor = torch.tensor([]) my_tensor = torch.tensor([[]]) my_tensor = torch.tensor([[[]]]) torch.max(input=my_tensor) # Error  my_tensor = torch.tensor([]) torch.max(input=my_tensor, dim=0) # Error  my_tensor = torch.tensor([[]]) torch.max(input=my_tensor, dim=0) # torch.return_types.max( # values=tensor([]), # indices=tensor([], dtype=torch.int64))  my_tensor = torch.tensor([[[]]]) torch.max(input=my_tensor, dim=0) # torch.return_types.max( # values=tensor([], size=(1, 0)), # indices=tensor([], size=(1, 0), dtype=torch.int64))
Enter fullscreen mode Exit fullscreen mode

Top comments (0)