Tensor multiplication 1D to 3D

vision
1

I was trying to multiply two tensors A - shape [torch.Size([1100, 4, 3])] with B - shape [torch.Size([1100])].
I would like to do this:

A[0]
--> 
tensor([[ 2.0025, -1.8551, -0.5907],
        [ 2.0805,  1.8512, -0.5358],
        [-1.9438, -2.0552, -0.5612],
        [-2.0561,  2.0529, -0.7012]], dtype=torch.float64)

B[0]
--> 
tensor(1.4853, dtype=torch.float64)

B[0] * A[0]
-->
tensor([[ 2.9742, -2.7553, -0.8773],
        [ 3.0902,  2.7496, -0.7958],
        [-2.8871, -3.0525, -0.8335],
        [-3.0538,  3.0491, -1.0415]], dtype=torch.float64)

Works if I choose some index, like above. But will not work if I use whole tensors. B * A ,

Error:
RuntimeError: The size of tensor a (1100) must match the size of tensor b (3) at non-singleton dimension 2

Not sure where I am missing?

2

This may help: Broadcasting semantics — PyTorch 1.12 documentation
That being said, it’d be best not to cross-post these things on the slack channel right after posting them on the forums :slight_smile:

3

Oops figured it out. converted shape of B to [1100,1,1]. Worked!!