I am trying to round the values in a 2D tensor to the nearest value from a given list, for example:
# this is eager, but the solution must work with non eager tensors
data_to_round = tf.constant([[0.3, 0.4, 2.3], [1.4, 2.2 ,55.4]])
possible_rounding_results = [1,2,3,4,5,6]
# TODO: round `data_to_round` to nearest values from `possible_rounding_results`
# expected output: [[1, 1, 2], [1, 2 , 5]]
I was playing around with tf.math.subtract, tf.math.abs and tf.argmin in order to find the index of the minimal absolute difference between the lists using a for loop, but then I failed to combine them back to a tensor, and it didn't work at all with 2D arrays only 1D.
And I am not sure if its the correct way to approach this problem at all.
Due to my total lack of experience with TensorFlow, I have no idea how to tackle such issue sorry in advance for the shallow question I am just lost.
If I understand you correctly, you could try something like this:
import tensorflow as tf
x = tf.constant([[0.3, 0.4, 2.5], [1.4, 2.2, 55.4]])
possible_rounding_results = tf.constant([1,2,3,4,5,6], dtype=tf.float32)
tf.clip_by_value(tf.round(x), tf.reduce_min(possible_rounding_results), tf.reduce_max(possible_rounding_results))
<tf.Tensor: shape=(2, 3), dtype=float32, numpy=
array([[1., 1., 2.],
[1., 2., 6.]], dtype=float32)>
The last value is 6 instead of 5 (as in your example) since 6 is closer to 55.4 than 5. If you meant to use tf.constant([[0.3, 0.4, 2.5], [1.4, 2.2, 5.4]]), you would get:
<tf.Tensor: shape=(2, 3), dtype=float32, numpy=
array([[1., 1., 2.],
[1., 2., 5.]], dtype=float32)>