[FIXED] Simplest solution to convert a 1D tensor to a 5x5 tensor

Issue

What would be the simplest solution to a convert the tensor:

tensor = tf.constant([4.0])

to a 5x5 tensor, where the main diagonal and the antidiagonal have the scalar 4.0, but the rest of the tensor has the value 0.0 as shown below:

tf.Tensor(
[[4. 0. 0. 0. 4.]
 [0. 4. 0. 4. 0.]
 [0. 0. 4. 0. 0.]
 [0. 4. 0. 4. 0.]
 [4. 0. 0. 0. 4.]], shape=(5, 5), dtype=float32)

Solution

Simply use tf.eye:

import tensorflow as tf

tensor = tf.where(tf.greater(tf.reverse(tf.eye(5), axis=[1]) + tf.eye(5), 0.0), 1.0, 0.0) * tf.constant([4.0])

tf.Tensor(
[[4. 0. 0. 0. 4.]
 [0. 4. 0. 4. 0.]
 [0. 0. 4. 0. 0.]
 [0. 4. 0. 4. 0.]
 [4. 0. 0. 0. 4.]], shape=(5, 5), dtype=float32)

Answered By - AloneTogether