Tensorflow - mapping python function with two parameters

Question

Hi I have a python function where I am trying to map a tensor to. I essentially need to run every element through the function. I am not sure how to map the two parameters to this function. Not only this but even when I remove the second parameter it gives me an error:

TypeError: bad operand type for unary -: 'list'

Here is my full code:

import tensorflow as tf

def sigmoid(x, derivative = False):
    if derivative == True:
        return (1.0/(1+math.exp(-x))) * (1.0 - (1.0/(1+math.exp(-x))))
    return 1.0/(1+math.exp(-x))

# build computational graph
a = tf.placeholder('float', None)

result = tf.map_fn(sigmoid, [a] , tf.float32)

# initialize variables
init = tf.global_variables_initializer()

# create session and run the graph
with tf.Session() as sess:
    sess.run(init)
    print sess.run(result, feed_dict={a: [2,3,4]})

# close session
sess.close()

I was following this: https://www.tensorflow.org/api_docs/python/tf/py_func

EDIT I can do the exp function using tensorflows library:

def sigmoid(x, derivative = False):
    if derivative == True:
        return (1.0/(1+tf.exp(-x))) * (1.0 - (1.0/(1+tf.exp(-x))))
    return 1.0/(1+tf.exp(-x))

Vijayachandran Mariappan · Accepted Answer · 2017-07-18 17:28:06Z

2

Why cant you use the tf.nn.sigmoid() function?.

def sigmoid(x, derivative = False):
   if derivative == True:
       return tf.nn.sigmoid(x) * (1.0 - tf.nn.sigmoid(x))
   return tf.nn.sigmoid(x)

If you want to call a numpy function in the graph, you can use tf.py_func (The code will be executed in CPU only):

def sigmoid(x, derivative = False):
if derivative == True:
    return (1.0/(1+np.exp(-x))) * (1.0 - (1.0/(1+np.exp(-x))))
return 1.0/(1+np.exp(-x))

# build computational graph
a = tf.placeholder('float', None)

result = tf.py_func(sigmoid, [a, True] , tf.float32)

edited Jul 18, 2017 at 17:28

answered Jul 17, 2017 at 22:55

Vijayachandran Mariappan

17.2k3 gold badges43 silver badges60 bronze badges

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

5 Comments

Kevin Over a year ago

This will work, but how can I get it to use the derivative? As in how can I pass two parameters into the function?

Vijayachandran Mariappan Over a year ago

Just call, result = sigmoid(a, derivative = True)?

Kevin Over a year ago

Thanks, How about if I needed to do an operation that required numpy. From what I understand I would not be able to do this?

Kevin Over a year ago

I thought that the py_func did this? tensorflow.org/api_docs/python/tf/py_func

Vijayachandran Mariappan Over a year ago

@kevin you are right, we can use py_func to call numpy functions. Updated the answer.

Collectives™ on Stack Overflow

Tensorflow - mapping python function with two parameters

1 Answer 1

5 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

5 Comments

Your Answer

Sign up or log in

Post as a guest

Related