Expected list for attr Tout: Tensorflow: while calling a python function while running the graph

Question

Well, I am stuck in a Tensor flow problem.

I am trying to call a python function from within tensorflow. According to the tensor flow manual py_func() can be used to invoke python function.

The want to achieve something like the below:

a = np.array([1,2,3,4], dtype='float32')
b = np.array([[5,6,7,8],[9,8,1,2], [3,2,3,1],[4,5,1,3]], dtype='float32')


def pyfunction(inputIN):
    return np.array(inputIN + a)

def tfFunction():
    inp = tf.placeholder(tf.float32, [2,4])
    out = tf.py_func(pyfunction, [inp], tf.float32)

tfFunction()   
with tf.Session() as sess:
    sess.run(tf.initialize_all_variables())    
    for inp_ in (b[0:2], b[2:4]):
        feed_dict = {inp:inp_}
        output = sess.run([out], feed_dict=feed_dict)
        print (output)

The output I require is :

[[  6.   8.  10.  12.]
 [ 10.  10.   4.   6.]]
[[ 4.  4.  6.  5.]
 [ 5.  7.  4.  7.]]

Using the above code I get an error.

TypeError: Expected list for attr Tout

I guess i get what the error says but I cant figure out a solution.

Please Note: I want to achieve a code very similar to it, The above problem is just a dummy problem. I am working on a image processing task and I have few image processing task (using OpenCV) inside a python function. I need to call the python function for every image while running the graph.

I understand that I can preprocess the data before and store it as batches, but I have few other tasks lined up. Therefore I have to stick to the above format

Any help will be appreciated. Thanks

S. Stas · Accepted Answer · 2017-04-04 10:44:58Z

2

This code provides the result desired:

import tensorflow as tf
import numpy as np

a = np.array([1,2,3,4], dtype='float32')
b = np.array([[5,6,7,8],[9,8,1,2], [3,2,3,1],[4,5,1,3]], dtype='float32')


def pyfunction(inputIN):
    return np.array(inputIN + a)

inp = tf.placeholder(tf.float32, [2,4])
out = tf.py_func(pyfunction, [inp], tf.float32)

with tf.Session() as sess:

    sess.run(tf.initialize_all_variables())
    for inp_ in (b[0:2], b[2:4]):
        feed_dict = {inp:inp_}
        output = sess.run(out, feed_dict=feed_dict)
        print (output)

Note, there are no square brackets near 'out' in the sess.run(..).

[[  6.   8.  10.  12.]
 [ 10.  10.   4.   6.]]
[[ 4.  4.  6.  5.]
 [ 5.  7.  4.  7.]]

answered Apr 4, 2017 at 10:44

S. Stas

8105 silver badges8 bronze badges

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

2 Comments

Sam Over a year ago

Sorry, it gives the same error even after using the code provided my you. Is it a version specific error, the version of tensor flow I am running is '0.10.0'

S. Stas Over a year ago

I used 1.x. It looks like it is a version-specific, really. So, for future cases, you should define version of library in the question description.

Collectives™ on Stack Overflow

Expected list for attr Tout: Tensorflow: while calling a python function while running the graph

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