当使用tf.cond（）时，Tensorflow报告'TypeError：需要单个张量时的张量列表'

Question

I am using Tensorflow to code a model. Part of my conditional statement like:

new_shape = tf.cond(tf.equal(tf.shape(src_shape)[0], 2), lambda: src_shape, lambda: tf.constant([1, src_shape[0]]))

and src_shape is the result of tf.shape() .

It reports TypeError: List of Tensors when single Tensor expected . I know it is because tf.constant([1, src_shape[0]]) is a list of tensors, but I don't know how to implement my code in a legal way.

I have try to remove tf.constant() like

new_shape = tf.cond(tf.equal(tf.shape(src_shape)[0], 2), lambda: src_shape, lambda: [1, src_shape[0]])

but it reports ValueError: Incompatible return values of true_fn and false_fn: The two structures don't have the same nested structure.

Answer 1

One way would be to use tf.stack, which stacks a list of rank-R tensors into one rank-(R+1) tensor.

lambda: tf.stack([1, src_shape[0]], axis=0)

Another solution would be using tf.concat using the right tf.reshape commands.

Answer 2

I have tried that tf.convert_to_tensor([1, src_shape[0]]) works. It is an alternative solution.