getting values in multiple indices from a tensor at once, in tensorflow

Question

Let's consider a numpy matrix, o:

If we want to use the following function by using numpy:

o[np.arange(x), column_array]

I can get multiple indices from a numpy array at once.

I have tried to do the same with tensorflow, but it doesn't work as what I have done. When o is a tensorflow tensor;

o[tf.range(0, x, 1), column_array]

I get the following error:

TypeError: can only concatenate list (not "int") to list

What can I do?

Answer 1

You can try tf.gather_nd(), as How to select rows from a 3-D Tensor in TensorFlow? this post suggested. Here is an example for getting multiple indices from matrix o.

o = tf.constant([[1, 2, 3, 4],
                 [5, 6, 7, 8],
                 [9, 10, 11, 12],
                 [13, 14, 15, 16]])
# [row_index, column_index], I don’t figure out how to
# combine row vector and column vector into this form.
indices = tf.constant([[0, 0], [0, 1], [2, 1], [2, 3]])

result = tf.gather_nd(o, indices)

with tf.Session() as sess:
    print(sess.run(result)) #[ 1  2 10 12]