Tensorflow session not executing function

Question

I am using Tensorflow placeholder in order to do a calculation however Tensorflow tells me that I have an initialization error.

import tensorflow as tf
import numpy as np

al = np.array([2.5, 0.0, 2, 8],dtype=float)
bl = np.array([3, -0.5, 2, 7],dtype=float)

sess = tf.Session()

a = tf.placeholder(dtype='float32')
b = tf.placeholder(dtype='float32')


c = tf.metrics.mean_squared_error(a,b)
d = sess.run(c, {a: al, b:bl } )

I cannot see where my mistake is. The error message is:

FailedPreconditionError (see above for traceback): Attempting to use uninitialized value mean_squared_error/count
     [[Node: mean_squared_error/count/read = Identity[T=DT_FLOAT, _class=["loc:@mean_squared_error/AssignAdd_1"], _device="/job:localhost/replica:0/task:0/device:CPU:0"](mean_squared_error/count)]]

Andreas Storvik Strauman · Accepted Answer · 2018-06-23 17:10:08Z

1

You have to initialise the variables that tf.metrics.mean_squared_error uses:

import tensorflow as tf
import numpy as np

a = tf.placeholder(dtype='float32')
b = tf.placeholder(dtype='float32')
c = tf.metrics.mean_squared_error(a,b)
sess = tf.Session()
# Initialise the local variables
init_local = tf.local_variables_initializer()
sess.run(init_local)
# Initialise the global variables (not really needed here)
# But you'll probably need it in the future when using tensorflow:
init_global = tf.global_variables_initializer()
sess.run(init_global)

al = np.array([2.5, 0.0, 2, 8],dtype=float)
bl = np.array([3, -0.5, 2, 7],dtype=float)
d = sess.run(c, {a: al, b:bl } )

Alternatively, in this case, you can use tf.losses.mean_squared_error.

I would check out tensorflows programmers guide on variables to get a better hold of these things :)

answered Jun 23, 2018 at 17:10

Andreas Storvik Strauman

1,66512 silver badges30 bronze badges

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

3 Comments

piccolo Over a year ago

Thank you. But how did you know that tf.metrics.mean_squared_error used local variables? I have also noticed that if I use tf.global_variables_initializer() on its own here I get the same error but if I use tf.local_variables_initializer() on its own the issue is resolved.

Andreas Storvik Strauman Over a year ago

@ImranRashid it wouldn't make sense to use variables that could be saved and restored internally in tf.metrics.mean_squared_error, so when it said that a variable inside there (count) was uninitialized then it was a fair chance it was local and not global. (Local variables won't get saved by a saver).

Andreas Storvik Strauman Over a year ago

@ImranRashid It also says in the docs: " The mean_squared_error function creates two local variables, total and count that are used to compute the mean squared error"

Vijayachandran Mariappan · Accepted Answer · 2018-06-23 17:27:50Z

1

FailedPreconditionError (see above for traceback): Attempting to use uninitialized value mean_squared_error/count

The value count from the above error, is one of the local variable metrics.mean_squared_error uses to calculate MSE (the other is count). So you need to initialize the local variables using:

sess.run(tf.local_variables_initializer())

edited Jun 23, 2018 at 17:27

answered Jun 23, 2018 at 17:09

Vijayachandran Mariappan

17.2k3 gold badges43 silver badges60 bronze badges

2 Comments

piccolo Over a year ago

Thanks Vijay. This worked as well but how did you know to use local_variables_initializer instead of global_variables_initializer?

Vijayachandran Mariappan Over a year ago

You dont need global_variables_initializer, the error is regarding local variables, i have added the link in my answer, there you can find the details of the two local variables.

Collectives™ on Stack Overflow

Tensorflow session not executing function

2 Answers 2

3 Comments

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

3 Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related