Swap tensor axes in TensorFlow without tf.transpose

Question

I have a tensor of unknown shape, but it is at least 3 dimensional, i.e. shape=[a, b, c, ...]. I would like to switch dimensions a and b, without knowing how long the tensor is (so I can't use tf.transpose, as suggested in this question)

did you find a better solution to this? – codekitty Feb 14 '21 at 22:58 — codekitty, Feb 14 '21 at 22:58

Toke Faurby · Answer 1 · 2017-06-29T11:49:04.103

2

This works, but is ugly:

tf.transpose(x, [1, 0] + [i+2 for i in range(tf.shape(x).shape[0]-2)])

edited Jun 29 '17 at 11:49

answered Jun 29 '17 at 11:38

Toke Faurby

5,788
9
41
62

2

You can use `tf.rank()` instead of `tf.shape()`: `tf.transpose(x, tf.concat([[1,0], tf.range(2, tf.rank(x))], 0)` – Manolo Santos Jun 29 '17 at 12:05

Swap tensor axes in TensorFlow without tf.transpose

1 Answers1