CODEWITHSUNDEEP
pythontensorflow
Marc Tim Thiemann
Marc Tim Thiemann

Reputation: 55

How do I evaluate a TensorFlow tuple?

import tensorflow as tf

a = tf.zeros([10])
b = tf.zeros([10])
state = tf.tuple([a, b], name='initial_state')

with tf.Session() as sess:
    s = sess.run('initial_state:0')

I get the following error with this example:

ValueError: Fetch argument 'initial_state' cannot be interpreted as a Tensor.
("The name 'initial_state' refers to an Operation not in the graph.")`

It works when I just pass the tensor, but not when I pass the name. Why can't I pass the name in this case?

Upvotes: 0

Views: 2360

Answers (2)

Crevious
Crevious

Reputation: 152

The answer of @E_net4 will behave is nice. But First, you should know how tf.tuple work. As in the tf.tuple documentation says 

tf.tuple( tensors, control_inputs=None, name=None )

You have to sometime follow the instruction. But it hard to understand there as no example shown so see mine :

import tensorflow as tf
a = tf.Variable([5])

b = tf.Variable([6])

c = a+b
d = a*b
e = a/b

ops = tf.tuple([c,d,e])
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    ee = sess.run(ops)
    print(ee)

Here ops = tf.tuple([tensor1,tensor2,...],control_inputs=c_ops)

The output shows :

[array([11], dtype=int32), array([30], dtype=int32), array([0.83333333])]`

Upvotes: 1

E_net4
E_net4

Reputation: 30042

Tuples in TensorFlow are not tensors, but a list of tensors, and so cannot be fetched as a whole through an operation in the graph. tf.tuple will create a few grouping and dependency control operations (initial_state/group_deps, initial_state/control_dependency and initial_state/control_dependency_1 in this case), but that's about it.

Since state is a list, it is a valid fetches argument to Session#run. One can also build a list of operation names from each tuple element and use that instead.

s = sess.run(['zeros:0', 'zeros_1:0'])
# [
#  array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32),
#  array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0.], dtype=float32)
# ]

Upvotes: 1

Related Questions