Tensorflow: How to index a tensor using 2D-index like in numpy

Question

I would like to do the following numpy code in Tensorflow:

input = np.array([[1,2,3]
                  [4,5,6]
                  [7,8,9]])
index1 = [0,1,2]
index2 = [2,2,0]
output = input[index1, index2]
>> output
[3,6,7]

given an input such as:

input = tf.constant([[1, 2, 3],
                     [4, 5, 6],
                     [7, 8, 9]])

I have tried the following, but seems like an overshooting:

index3 = tf.range(0, input.get_shape()[0])*input.get_shape()[1] + index2
output = tf.gather(tf.reshape(input, [-1]), index3)
sess = tf.Session()
sess.run(output)
>> [3,6,7]

This works only because my first index is conveniently [0,1,2] but wouldn't be doable for [0,0,2] for example (besides looking really long and ugly).

Would you have any easier syntax, something more tensoric/pythonic?

Answer 1

You can do it using tf.gather_nd(tf.gather_nd official doc) as follows:

import tensorflow as tf
inp = tf.constant([[1, 2, 3],
                   [4, 5, 6],
                   [7, 8, 9]])
res=tf.gather_nd(inp,list(zip([0,1,2],[2,2,0])))
sess = tf.Session()
sess.run(res)

The result is array([3, 6, 7])