Tensors with same name in the python code in tensorflow

Question

According to the code in: https://github.com/tensorflow/models/blob/master/tutorials/image/cifar10/cifar10.py, it happens that the same names are used for the tensors variables such as:

conv = tf.nn.conv2d(images, kernel, [1, 1, 1, 1], padding='SAME') # Under conv1, line: 208

and,

conv = tf.nn.conv2d(norm1, kernel, [1, 1, 1, 1], padding='SAME') # Under conv2, line 227

Therefore, why this is allowed in tensorflow? If for some reason, If I tried to say:

sess.run([conv], feed_dict{x: some_data})

Then which conv tensor we will be evaluated?

Second, if the conv tensor under CONV1 layer was referring to the tf.nn.conv2d operation. How could another conv tensor under CONV2 refer to the second tf.nn.conv2d operation? In other words, how they are treated separately?

Any help is much appreciated!!

I think, all of those are under different namescopes. Their TF names are different. So, they don't conflict — Harsha Pokkalla
– Harsha Pokkalla, Commented Jun 9, 2017 at 18:37
@hars, Yea I totally agree with you. But from the coding perspective, how is that possible? — I. A
– I. A, Commented Jun 9, 2017 at 19:28

Harsha Pokkalla · Accepted Answer · 2017-06-09 20:02:08Z

2

for your question: latest "conv" is evaluated

For example:

import tensorflow as tf

a = tf.constant(5)
b = tf.constant(6)

c = tf.multiply(a,b)
print c
c = tf.multiply(c,b)
print c

sess = tf.Session()
c_val = sess.run(c)
print c_val

Output :

Tensor("Mul:0", shape=(), dtype=int32)
Tensor("Mul_1:0", shape=(), dtype=int32)
180

You can see TF names them differently. Whenever you call an TF operator, it creates a node independent of python variable name. But python variable name corresponds to latest tensor you used.

I hope this helps.

answered Jun 9, 2017 at 20:02

Harsha Pokkalla

1,8121 gold badge13 silver badges17 bronze badges

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

Collectives™ on Stack Overflow

Tensors with same name in the python code in tensorflow

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related