DEV Community

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

Posted on • Edited on

flatten and ravel in PyTorch

#python #pytorch #flatten #ravel

Buy Me a Coffee☕

*Memos:

flatten() can remove zero or more dimensions by selecting dimensions from the 0D or more D tensor of zero or more elements, getting the 1D or more D tensor of zero or more elements as shown below:

*Memos:

  • flatten() 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 start_dim(Optional-Default:0-Type:int).
  • The 3rd argument with torch or the 2nd argument with a tensor is end_dim(Optional-Default:-1-Type:int).
  • flatten() can change a 0D tensor to a 1D tensor.
  • flatten() does nothing for a 1D tensor.
  • The difference between Flatten() and flatten() is:
    • The default value of start_dim for Flatten() is 1 while the default value of start_dim for flatten() is 0.
    • Basically, Flatten() is used to define a model while flatten() is not used to define a model. 
import torch my_tensor = torch.tensor(7) torch.flatten(input=my_tensor) my_tensor.flatten() torch.flatten(input=my_tensor, start_dim=0, end_dim=0) torch.flatten(input=my_tensor, start_dim=0, end_dim=-1) torch.flatten(input=my_tensor, start_dim=-1, end_dim=0) torch.flatten(input=my_tensor, start_dim=-1, end_dim=-1) # tensor([7])  my_tensor = torch.tensor([7, 1, -8, 3, -6, 0]) torch.flatten(input=my_tensor) torch.flatten(input=my_tensor, start_dim=0, end_dim=0) torch.flatten(input=my_tensor, start_dim=0, end_dim=-1) torch.flatten(input=my_tensor, start_dim=-1, end_dim=0) torch.flatten(input=my_tensor, start_dim=-1, end_dim=-1) # tensor([7, 1, -8, 3, -6, 0])  my_tensor = torch.tensor([[7, 1, -8], [3, -6, 0]]) torch.flatten(input=my_tensor) torch.flatten(input=my_tensor, start_dim=0, end_dim=1) torch.flatten(input=my_tensor, start_dim=0, end_dim=-1) torch.flatten(input=my_tensor, start_dim=-2, end_dim=1) torch.flatten(input=my_tensor, start_dim=-2, end_dim=-1) # tensor([7, 1, -8, 3, -6, 0])  torch.flatten(input=my_tensor, start_dim=0, end_dim=0) torch.flatten(input=my_tensor, start_dim=-1, end_dim=-1) torch.flatten(input=my_tensor, start_dim=0, end_dim=-2) torch.flatten(input=my_tensor, start_dim=1, end_dim=1) torch.flatten(input=my_tensor, start_dim=1, end_dim=-1) torch.flatten(input=my_tensor, start_dim=-1, end_dim=1) torch.flatten(input=my_tensor, start_dim=-1, end_dim=-1) torch.flatten(input=my_tensor, start_dim=-2, end_dim=0) torch.flatten(input=my_tensor, start_dim=-2, end_dim=-2) # tensor([[7, 1, -8], [3, -6, 0]])  my_tensor = torch.tensor([[[7], [1], [-8]], [[3], [-6], [0]]]) torch.flatten(input=my_tensor) torch.flatten(input=my_tensor, start_dim=0, end_dim=2) torch.flatten(input=my_tensor, start_dim=0, end_dim=-1) torch.flatten(input=my_tensor, start_dim=-3, end_dim=2) torch.flatten(input=my_tensor, start_dim=-3, end_dim=-1) # tensor([7, 1, -8, 3, -6, 0])  torch.flatten(input=my_tensor, start_dim=0, end_dim=0) torch.flatten(input=my_tensor, start_dim=0, end_dim=-3) torch.flatten(input=my_tensor, start_dim=1, end_dim=1) torch.flatten(input=my_tensor, start_dim=1, end_dim=-2) torch.flatten(input=my_tensor, start_dim=2, end_dim=2) torch.flatten(input=my_tensor, start_dim=2, end_dim=-1) torch.flatten(input=my_tensor, start_dim=-1, end_dim=2) torch.flatten(input=my_tensor, start_dim=-1, end_dim=-1) torch.flatten(input=my_tensor, start_dim=-2, end_dim=1) torch.flatten(input=my_tensor, start_dim=-2, end_dim=-2) torch.flatten(input=my_tensor, start_dim=-3, end_dim=0) torch.flatten(input=my_tensor, start_dim=-3, end_dim=-3) # tensor([[[7], [1], [-8]], [[3], [-6], [0]]])  torch.flatten(input=my_tensor, start_dim=0, end_dim=1) torch.flatten(input=my_tensor, start_dim=0, end_dim=-2) torch.flatten(input=my_tensor, start_dim=-3, end_dim=1) torch.flatten(input=my_tensor, start_dim=-3, end_dim=-2) # tensor([[7], [1], [-8], [3], [-6], [0]])  torch.flatten(input=my_tensor, start_dim=1, end_dim=2) torch.flatten(input=my_tensor, start_dim=1, end_dim=-1) torch.flatten(input=my_tensor, start_dim=-2, end_dim=2) torch.flatten(input=my_tensor, start_dim=-2, end_dim=-1) # tensor([[7, 1, -8], [3, -6, 0]])  my_tensor = torch.tensor([[[7.], [1.], [-8.]], [[3.], [-6.], [0.]]]) torch.flatten(input=my_tensor) # tensor([7., 1., -8., 3., -6., 0.])  my_tensor = torch.tensor([[[7.+0.j], [1.+0.j], [-8.+0.j]], [[3.+0.j], [-6.+0.j], [0.+0.j]]]) torch.flatten(input=my_tensor) # tensor([7.+0.j, 1.+0.j, -8.+0.j, 3.+0.j, -6.+0.j, 0.+0.j])  my_tensor = torch.tensor([[[True], [False], [True]], [[False], [True], [False]]]) torch.flatten(input=my_tensor) # tensor([True, False, True, False, True, False])
Enter fullscreen mode Exit fullscreen mode

ravel() can remove zero or more dimensions as much as possible from the 0D or more D tensor of zero or more elements, getting the 1D tensor of zero or more elements as shown below:

*Memos:

  • ravel() 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).
  • ravel() can change a 0D tensor to a 1D tensor.
  • ravel() does nothing for a 1D tensor. 
import torch my_tensor = torch.tensor(7) torch.ravel(input=my_tensor) my_tensor.ravel() # tensor([7])  my_tensor = torch.tensor([7, 1, -8, 3, -6, 0]) torch.ravel(input=my_tensor) # tensor([7, 1, -8, 3, -6, 0])  my_tensor = torch.tensor([[7, 1, -8], [3, -6, 0]]) torch.ravel(input=my_tensor) # tensor([7, 1, -8, 3, -6, 0])  my_tensor = torch.tensor([[[7], [1], [-8]], [[3], [-6], [0]]]) torch.ravel(input=my_tensor) # tensor([7, 1, -8, 3, -6, 0])  my_tensor = torch.tensor([[[7.], [1.], [-8.]], [[3.], [-6.], [0.]]]) torch.ravel(input=my_tensor) # tensor([7., 1., -8., 3., -6., 0.])  my_tensor = torch.tensor([[[7.+0.j], [1.+0.j], [-8.+0.j]], [[3.+0.j], [-6.+0.j], [0.+0.j]]]) torch.ravel(input=my_tensor) # tensor([7.+0.j, 1.+0.j, -8.+0.j, 3.+0.j, -6.+0.j, 0.+0.j])  my_tensor = torch.tensor([[[True], [False], [True]], [[False], [True], [False]]]) torch.ravel(input=my_tensor) # tensor([True, False, True, False, True, False])
Enter fullscreen mode Exit fullscreen mode

Top comments (0)