Element wise multiplication of the last dimension

1

Hello everyone,

I have the following two tensors:

E with size (3, 2, 4), lets visualize it as
e11, e12
e21, e22
e31, e32
where every e is a vector with dimension 4

G with size (3, 2), lets visualize it as
g11, g12
g21, g22
g31, g32
where every g is a scalar

As a result I want to multiply every vector e of E with the corresponding scalar g from G as following:
g11*e11, g12*e12
g21*e21, g22*e22
g31*e31, g32*e32
where * is element wise multiplication, or in other words every vector e from the tensor E (let’s say e11), I want it multiplied with the corresponding scalar (or g11 for e11), etc… The size of the result should be (2, 2, 4), same as the E tensor.

Thank you.

1 Like
2

I found out that first unsqueezing the G tensor, repeating it 4 times along the 3-th dimension, and element-wise multiplying it with E does the job, but there may be a more elegant solution. Here is the code:

G_tmp = G.unsqueeze(2).expand(-1, -1, 4)
res = G_tmp * E

Feel free to correct me, or propose a more elegant solution :grin:

3

You don’t need expand() there, tensor broadcasting takes care of it.