torch.stack

[ShreyasMuragodmath]

I tried torch.stack with dim = 0 and dim = 1.
torch.vstack is same as torch,stack with dim=0
torch.hstack just concatenates and extends columns, and and is not the same as torch.stack with dim=1
as mentioned in the tutorial

x = torch.arange(1., 10.)
x, x.shape

(tensor([1., 2., 3., 4., 5., 6., 7., 8., 9.]), torch.Size([9]))

x_stacked = torch.stack([x, x, x, x], dim=0)
x, x.shape, x_stacked, x_stacked.shape

(tensor([7., 2., 3., 4., 5., 6., 7., 8., 9.]),
 torch.Size([9]),
 tensor([[7., 2., 3., 4., 5., 6., 7., 8., 9.],
         [7., 2., 3., 4., 5., 6., 7., 8., 9.],
         [7., 2., 3., 4., 5., 6., 7., 8., 9.],
         [7., 2., 3., 4., 5., 6., 7., 8., 9.]]),
 torch.Size([4, 9]))

x_stacked = torch.stack([x, x, x, x], dim=1)
x, x.shape, x_stacked, x_stacked.shape

(tensor([7., 2., 3., 4., 5., 6., 7., 8., 9.]),
 torch.Size([9]),
 tensor([[7., 7., 7., 7.],
         [2., 2., 2., 2.],
         [3., 3., 3., 3.],
         [4., 4., 4., 4.],
         [5., 5., 5., 5.],
         [6., 6., 6., 6.],
         [7., 7., 7., 7.],
         [8., 8., 8., 8.],
         [9., 9., 9., 9.]]),
 torch.Size([9, 4]))

x_hstacked = torch.hstack([x, x, x, x])
x, x.shape, x_hstacked, x_hstacked.shape

(tensor([7., 2., 3., 4., 5., 6., 7., 8., 9.]),
 torch.Size([9]),
 tensor([7., 2., 3., 4., 5., 6., 7., 8., 9., 7., 2., 3., 4., 5., 6., 7., 8., 9.,
         7., 2., 3., 4., 5., 6., 7., 8., 9., 7., 2., 3., 4., 5., 6., 7., 8., 9.]),
 torch.Size([36]))

x_vstacked = torch.vstack([x, x, x, x])
x, x.shape, x_vstacked, x_vstacked.shape

(tensor([7., 2., 3., 4., 5., 6., 7., 8., 9.]),
 torch.Size([9]),
 tensor([[7., 2., 3., 4., 5., 6., 7., 8., 9.],
         [7., 2., 3., 4., 5., 6., 7., 8., 9.],
         [7., 2., 3., 4., 5., 6., 7., 8., 9.],
         [7., 2., 3., 4., 5., 6., 7., 8., 9.]]),
 torch.Size([4, 9]))

[Haji-Suleman]

Your observations are correct, and there’s a small clarification needed in the tutorial wording.

Key Difference

torch.stack creates a new dimension

torch.hstack / torch.vstack concatenate along an existing dimension

Correct Equivalences

torch.vstack([x, x, x, x]) ✅
↔ torch.stack([x, x, x, x], dim=0) for 1D tensors

torch.hstack([x, x, x, x]) ❌
≠ torch.stack(..., dim=1)

Why?

stack(dim=1) → shape (9, 4) (new axis)

hstack → shape (36,) (just concatenation)

So the tutorial statement comparing hstack with stack(dim=1) is misleading.
hstack is closer to torch.cat(..., dim=0) for 1D tensors.

[mrdbourke]

Hi there,

Thank you for the clarifications, can you please link/share where in the tutorials needs updating?