Scalar matrix multiplication for a tensor and an array of scalars

1

If I have a 3d tensor of size
T1 --> torch.Size([196, 14, 14])
and a 1d tensor of size
T2 --> torch.Size([196])

How do I multiply matrices of size [14, 14] in T1 by the corresponding scalar in T2?

e.g if T2[0] == 3, then the first multiplication would be
T1[0] * 3 , etc.

So the output should be with the original shape of T1.

2

Hi,

You just need to change the view of second tensor to match shapes in corresponding dimension, then PyTorch’s tensor broadcasting will take care of that. e.g.

t1 = torch.arange(0, 196*14*14).view((196, 14, 14))  # shape (196, 14, 14)
t2 = torch.arange(196).view(-1, 1, 1)  # shape (196, 1, 1)
result = t1*t2

Bests

1 Like