Tensorflow如何获得张量值

Question

In python I can use

a = np.array([[3], [6], [9]])

Obviously,

a[0][0] = 3
a[1][0] = 6
a[2][0] = 9

But I tried to do the same thing with tensorflow

import tensorflow as tf
a = tf.Variable(np.array([[3], [6], [9]]))
init = tf.initialize_all_variables()

with tf.Session() as ss:
   ss.run(init)
   for i in range(3):
       print sess.run(a[i][0])

If I print it(use for loop), I got TypeError: 'Variable' object is not callable

How can I resolve this error? Thanks very much for any help!

Answer 1

You can define another op, that is dependent on the original variable, that contains the slice of your tensor:

import tensorflow as tf
a = tf.Variable(np.array([[3], [6], [9]]))
part = []
for i in range(3):
    part.append(a[i][0])
init = tf.initialize_all_variables()

with tf.Session() as ss:
   ss.run(init)
   for op in part:
       print ss.run(op)

Answer 2

Despite tensorflow and numpy are quite similar at the first glance, tensorflow workflow substantially differs from numpy's. When using tensorflow, you should first define the computational graph -- the rules defining the connections between tensors.

In your case, the graph consists of only one variable a . Once the graph was defined, you would be able to compute the values of different nodes in a graph by running a tensorflow session. In your case, to print a value of a , use the following code:

sess = tf.Session()
init = tf.initialize_all_variables()
sess.run(init)
print(sess.run(a))