DEV Community

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

Posted on • Edited on

dim, size, item and tolist in PyTorch

Buy Me a Coffee

*My post explains how to create and acceess a tensor.

dim() can get the number of dimensions from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

  • dim() can be used with a tensor but not with torch.
  • ndim is the alias of dim().
import torch my_tensor = torch.tensor(-3) # 0D tensor  my_tensor.dim() # 0  my_tensor = torch.tensor([-3, 7, -5, 2, 6, 3]) # 1D tensor my_tensor.dim() # 1 my_tensor[0].dim() # 0  my_tensor = torch.tensor([[-3, 7, -5], [2, 6, 3], # 2D tensor  [8, 0, -1], [4, 9, -6]]) my_tensor.dim() # 2 my_tensor[0].dim() # 1 my_tensor[0][0].dim() # 0  my_tensor = torch.tensor([[[-3, 7], [-5, 2], [6, 3]], # 3D tensor  [[8, 0], [-1, 4], [9, -6]], [[5, -2], [-7, 9], [1, 1]], [[6, -4], [0, -9], [3, 5]]]) my_tensor.dim() # 3 my_tensor[0].dim() # 2 my_tensor[0][0].dim() # 1 my_tensor[0][0][0].ndim # 0 
Enter fullscreen mode Exit fullscreen mode

size() can get a size from the 0D or more D tensor of zero or more elements as shown below:

*Memos:

  • size() can be used with a tensor but not with torch.
  • The 1st argument with a tensor is dim(Optional-Type:int).
  • shape is the alias of dim().
import torch my_tensor = torch.tensor(-3) # 0D tensor  my_tensor.size() # torch.Size([])  my_tensor = torch.tensor([-3, 7, -5, 2, 6, 3]) # 1D tensor  my_tensor.size() # torch.Size([6])  my_tensor.size(dim=0) my_tensor.size(dim=-1) # 6  my_tensor[0].size() # torch.Size([])  my_tensor = torch.tensor([[-3, 7, -5], [2, 6, 3], # 2D tensor  [8, 0, -1], [4, 9, -6]]) my_tensor.size() # torch.Size([4, 3])  my_tensor.size(dim=0) my_tensor.size(dim=-2) # 4  my_tensor.size(dim=1) my_tensor.size(dim=-1) my_tensor[0].size(dim=0) my_tensor[0].size(dim=-1) # 3  my_tensor[0].size() # torch.Size([3])  my_tensor[0][0].size() # torch.Size([])  my_tensor = torch.tensor([[[-3, 7], [-5, 2], [6, 3]], # 3D tensor  [[8, 0], [-1, 4], [9, -6]], [[5, -2], [-7, 9], [1, 1]], [[6, -4], [0, -9], [3, 5]]]) my_tensor.size() # torch.Size([4, 3, 2])  my_tensor.size(dim=0) my_tensor.size(dim=-3) # 4  my_tensor.size(dim=1) my_tensor.size(dim=-2) my_tensor[0].size(dim=0) my_tensor[0].size(dim=-2) # 3  my_tensor.size(dim=2) my_tensor.size(dim=-1) my_tensor[0].size(dim=1) my_tensor[0].size(dim=-1) my_tensor[0][0].size(dim=0) my_tensor[0][0].size(dim=-1) # 2  my_tensor[0].size() # torch.Size([3, 2])  my_tensor[0][0].size() # torch.Size([2])  my_tensor[0][0][0].size() # torch.Size([]) 
Enter fullscreen mode Exit fullscreen mode

item() can get a standard Python number from the 0D or more D tensor of only one element as shown below:

*Memos:

  • item() can be used with a tensor but not with torch.
  • A tensor must be a scalar.
import torch my_tensor = torch.tensor(-3) # 0D tensor my_tensor = torch.tensor([-3]) # 1D tensor my_tensor = torch.tensor([[-3]]) # 2D tensor my_tensor = torch.tensor([[[-3]]]) # 3D tensor  my_tensor.item() # -3 
Enter fullscreen mode Exit fullscreen mode

tolist() can get a standard Python number or list from the 0D or more D tensor of zero or more elements as shown below. *tolist() can be used with a tensor but not with torch:

import torch my_tensor = torch.tensor(-3) # 0D tensor  my_tensor.tolist() # -3  my_tensor = torch.tensor([-3, 7, -5, 2, 6, 3]) # 1D tensor  my_tensor.tolist() # [-3, 7, -5, 2, 6, 3]  my_tensor[0].tolist() # -3  my_tensor = torch.tensor([[-3, 7, -5], [2, 6, 3], # 2D tensor  [8, 0, -1], [4, 9, -6]]) my_tensor.tolist() # [[-3, 7, -5], [2, 6, 3], # [8, 0, -1], [4, 9, -6]]  my_tensor[0].tolist() # [-3, 7, -5]  my_tensor[0][0].tolist() # -3  my_tensor = torch.tensor([[[-3, 7], [-5, 2], [6, 3]], # 3D tensor  [[8, 0], [-1, 4], [9, -6]], [[5, -2], [-7, 9], [1, 1]], [[6, -4], [0, -9], [3, 5]]]) my_tensor.tolist() # [[[-3, 7], [-5, 2], [6, 3]], # [[8, 0], [-1, 4], [9, -6]], # [[5, -2], [-7, 9], [1, 1]], # [[6, -4], [0, -9], [3, 5]]]  my_tensor[0].tolist() # [[-3, 7], [-5, 2], [6, 3]]  my_tensor[0][0].tolist() # [-3, 7]  my_tensor[0][0][0].tolist() # -3 
Enter fullscreen mode Exit fullscreen mode

Top comments (0)