用于张量乘法的 keras/tensorflow 广播

Question

I have two tensors: tensor a has shape (n, 1) , and tensor b has shape (p, q) where both p and q are small but n is large. I would like to get their "outer product" tensor c as a tensor with shape (n, p, q) , such that its elements are

c[i, j, k] = a[i, 0] * b[j, k]

How should I use tensorflow/keras broadcast or functions to achieve this without incurring a loop?

Answer 1

你可以简单地做：

c = tf.expand_dims(a, 2) * b