3d乘2d矩阵Tensorflow

Question

Suppose I have two tensors of different dimensions.

<tf.Tensor 'embedding_lookup:0' shape=(?, 128) dtype=float32>
<tf.Tensor 'Neg:0' shape=(?, 15, 128) dtype=float32>

I want to do a matrix multiplication of these two tensors such that I get a final tensor with shape (?, 15, 1) or (?, 15) . How could I do this?

Answer 1

With X and Y as the 2D and 3D inputs respectively, we can use tf.einsum -

tf.einsum('ij,ikj->ik',X,Y)

The output would have a shape of (?, 15) .