CODEWITHSUNDEEP
pythonnumpytensorflow
yusuf
yusuf

Reputation: 3781

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

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?

Upvotes: 5

Views: 5142

Answers (2)

Hui.X
Hui.X

Reputation: 136

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]

Upvotes: 5

Qy Zuo
Qy Zuo

Reputation: 2732

You may want to see tf.gather_nd: https://www.tensorflow.org/api_docs/python/tf/gather_nd

import tensorflow as tf
import numpy as np

tensor = tf.placeholder(tf.float32, [2,2])
indices = tf.placeholder(tf.int32, [2,2])
selected = tf.gather_nd(tensor, indices=indices)

with tf.Session() as session:
    data = np.array([[0.1,0.2],[0.3,0.4]])
    idx = np.array([[0,0],[1,1]])
    result = session.run(selected, feed_dict={indices:idx, tensor:data})
    print(result)

and the result will be [ 0.1 0.40000001]

Upvotes: 1

Related Questions