variable length for loop - tensorflow

Question

I want to do for loop's with dynamic inputs in tensorflow.

I know for the code below reduce_sum is sufficient, but I want to be able to formulate code in the manner described below. So given that I have an array with different sizes, I want to traverse that array.

How can I traverse dynamic length arrays?

Problem:

TypeError: range() integer end argument expected, got Tensor

How can I evaluate get_sum that is outside the session?

sum = get_sum(x, xshape) shouldn't be in the for loop inside session

For example:

import tensorflow as tf
import numpy as np


x = tf.placeholder(tf.float32, shape=[None])
xshape = tf.placeholder(tf.float32, shape=[])


def get_sum(x, xshape):
  sum = 0
  for i in range(xshape):
    sum += x[i]

init = tf.initialize_all_variables()
sum = get_sum(x, xshape)

with tf.Session() as sess:
  sess.run(init)
  for i in range(100):
    length = np.random.randint(0,10)
    a = np.random.randint(0, 10, length)
  print sess.run(sum,feed_dict={x:a, xshape:length})

If the number of iterations is unknown ahead of time, wouldn't another kind of loop be more appropriate? — Scott Hunter
– Scott Hunter, Commented Mar 13, 2016 at 15:40
it is the same problem, you can't get shape as an integer; how are you going to use while loop?- the problem is to convert the shape into an integer, we need to evaluate it; the problem is how? — user2776945
– user2776945, Commented Mar 13, 2016 at 15:43
So your question has nothing to do with a loop, but rather how to get a certain integer: got it. — Scott Hunter
– Scott Hunter, Commented Mar 13, 2016 at 15:46

Lukasz Kaiser · Accepted Answer · 2016-03-18 03:48:06Z

1

Dynamic loops are still young in TensorFlow, but they exist. Take a look for example at the fold operators and their implementation. https://www.tensorflow.org/versions/master/api_docs/python/control_flow_ops.html#foldl

answered Mar 18, 2016 at 3:48

Lukasz Kaiser

2,19614 silver badges4 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

CoderOnly Over a year ago

What is the situation now? Thank you

Lukasz Kaiser Over a year ago

Dynamic loops are well-supported now and there is a number of higher-level operators to make using them easier. You can do, e.g., elems = [1, 2, 3, 4, 5, 6] ; sum = tf.foldl(lambda a, x: a + x, elems), check out tf.foldl and tf.scan API docs. tensorflow.org/api_docs/python/tf/foldl

Collectives™ on Stack Overflow

variable length for loop - tensorflow

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related