How to multiply tensors of different shapes? (broadcasting)

Question

I want to multiple the elements in a tensor of shape [a,b,c], each in the 3rd dimension with a scalar from a tensor of shape [a,b].

For example,

x = 
|[1,2][3,4][5,6]|
|[1,2][3,4][5,6]|
|[1,2][3,4][5,6]|

y = 
|0 0 1|
|1 0 0|
|0 1 0|

I want to multiply x and y to obtain:

res = 
|[0,0][0,0][5,6]|
|[1,2][0,0][0,0]|
|[0,0][3,4][0,0]|

Could you tell me any method to achieve this?

P.S: maybe I can make a y' from y which is

y' =
|[0,0][0,0][1,1]|
|[1,1][0,0][0,0]|
|[0,0][1,1][0,0]|

then I can use tf.mul() to obtain the result. But I don't find anything to duplicate elements in a tensor in this way.

Answer 1

You are right, you can just use tf.mul(). In fact, the multiply operation in TensorFlow handles broadcasting, so you don't need to create another vector of shape [a, b, c].

To make sure that the broadcasting is on the expected dimension, you can add a third dimension to your second vector y:

x = tf.random_normal([3, 3, 2])
y = tf.constant([[0., 0., 1.], [1., 0., 0.], [0., 1., 0.]])

y = tf.expand_dims(y, 2)  # y will now have a matching shape of [3, 3, 1]
res = tf.mul(x, y)

sess = tf.Session()
sess.run(res)