如何在 pytorch 中连接两个不同尺寸的张量

Question

I have two tensors in pytorch with these shapes:

torch.Size([64, 100]) and torch.Size([64, 100, 256])

I want to concate them by torch.cat but they should be in the same shape and size. So I get this error:

RuntimeError: Tensors must have same number of dimensions: got 2 and 3

what should I do to fix this problem? how can I convert 2d PyTorch tensor into 3d tensor OR how can I convert 3d PyTorch tensor to 2d tensor without losing any data? or any other idea?

Answer 1

Depending on what you are looking to do with those two tensors, you could consider concatenating on the last axis such that the resulting tensor is shaped (64, 100, 257) . This requires you first unsqueeze a singleton dimensions on the first tensor:

>>> x, y = torch.rand(64, 100), torch.rand(64, 100, 256)
>>> z = torch.cat((x[..., None], y), -1)

>>> z.shape
torch.Size([64, 100, 257])