beginner pytorch tensor-cheatsheet.

tensor_cheatsheet.md

Pytorch Tensor Cheatsheet

1. Initialization

• Scalar (0D tensor):

  x = torch.tensor(42)
  # Shape: []

• Vector (1D tensor):

  x = torch.tensor([1, 2, 3])
  # Shape: [3]

• Matrix (2D tensor):

  x = torch.tensor([[1, 2], [3, 4], [5, 6]])
  # Shape: [3, 2]

• 3D tensor:

  x = torch.tensor([[[1, 2], [3, 4]], [[5, 6], [7, 8]]])
  # Shape: [2, 2, 2]

2. Reshaping

• Reshape/View:

  x = torch.randn(2, 3)
  y = x.view(6)
  # Shape of y: [6]

• Squeeze (remove dimensions of size 1):

  x = torch.randn(1, 3, 1)
  y = x.squeeze()
  # Shape of y: [3]

• Unsqueeze (add a dimension of size 1):

  x = torch.randn(3)
  y = x.unsqueeze(0)
  # Shape of y: [1, 3]

3. Combining Tensors

• Stack:

  x = torch.tensor([1, 2])
  y = torch.tensor([3, 4])
  z = torch.stack((x, y))
  # Shape of z: [2, 2]

• Concatenate:

  x = torch.tensor([[1, 2]])
  y = torch.tensor([[3, 4]])
  z = torch.cat((x, y), dim=0)
  # Shape of z: [2, 2]

4. Reduction Operations

• Sum along a dimension:

  x = torch.tensor([[1, 2, 3], [4, 5, 6]])
  y = torch.sum(x, dim=0)
  # Shape of y: [3]

5. Expansion/Broadcasting

• Expand dimensions:

  x = torch.randn(3)
  y = x[:, None]  # Adds a new dimension
  # Shape of y: [3, 1]