如何在 tensorflow 中连接两个具有动态形状的张量？

Question

I have two tensor I want to contact in tensorflow.

So the shape of these tensor are: (Bs,dynamics,n_features)

The batch_size and n_features are guarantee to be same, so I think this operation is possible.

However, I think tensorflow will run a pre-check? (sorry I am new to tensorflow), tensorflow will check if two tensor with shape (None,None,n_features) could be concatenated in axis 1, which is not possible, because it doesn't know what is the result of None+None

Anybody has any idea to this?

Answer 1

Concatenating should not be a problem in your case:

import tensorflow as tf

x1 = tf.random.normal((5, 10, 20))
x2 = tf.random.normal((5, 33, 20))

@tf.function
def concatenate(x1, x2):
  return tf.concat([x1, x2], axis=1)

concatenate(x1, x2).shape