InvalidArgumentError: unique expects a 1D vector

Question

I\'m trying to create a customized loss function to use in a neural network algorithm. My loss function:

import keras.backend as kb
import tensorflow as tf
def si