torch.vstack¶
- torch.vstack(tensors, *, out=None) Tensor ¶
Stack tensors in sequence vertically (row wise).
This is equivalent to concatenation along the first axis after all 1-D tensors have been reshaped by
torch.atleast_2d()
.- Parameters:
tensors (sequence of Tensors) – sequence of tensors to concatenate
- Keyword Arguments:
out (Tensor, optional) – the output tensor.
Example:
>>> a = torch.tensor([1, 2, 3]) >>> b = torch.tensor([4, 5, 6]) >>> torch.vstack((a,b)) tensor([[1, 2, 3], [4, 5, 6]]) >>> a = torch.tensor([[1],[2],[3]]) >>> b = torch.tensor([[4],[5],[6]]) >>> torch.vstack((a,b)) tensor([[1], [2], [3], [4], [5], [6]])