Pfthb

I know there are related questions out there, and I have absolutely looked at them... but for some reason it's not making sense to me, so I need to ask for some help here, as I have spent too long on this. I've written a simple code example. Quite simply I have a placeholder value, which I want to pass to a function which creates some tensorflow ops... then later I want to call another function which is able to access the values from those ops, without passing around actual variables. My goal was to access these variables via custom collection names. My example defines variables via "z1" = tf.Variable and "z2" = tf.get_variable, as well as assigning the calculation directly to "y". I'm obviously missing something very basic, as the only value which I see dynamically changing is in "y", where I have directly assigned the value.

import tensorflow as tf
import numpy as np

x = tf.placeholder(tf.float32, shape=(None, )+ (12,))
z1 = tf.Variable(0.0, dtype=tf.float32,
 validate_shape=False,
 trainable=False,
 collections = [
 tf.GraphKeys.GLOBAL_VARIABLES,
 "scale"
 ])
z2 = tf.get_variable("z2", , tf.float32,
 tf.initializers.zeros(), 
 collections = [
 tf.GraphKeys.GLOBAL_VARIABLES,
 "scale" 
 ],
 trainable=False)

def build_graph(x):
 z1_op = tf.assign(z1, tf.reduce_mean(x))
 z2_op = tf.assign(z2, tf.reduce_mean(x))
 y = tf.reduce_mean(x)
 return y

def rescale():
 q = tf.get_collection("scale")
 return q

def global_vars():
 q = tf.global_variables()
 return q

y = build_graph(x)
q1 = rescale()
q2 = global_vars()

data = np.random.rand(500, 12)

init_op = tf.global_variables_initializer()
with tf.Session() as sess:

 sess.run(init_op)
 for n in range(5):
 batch_idx = np.random.choice(list(range(500)), np.random.randint(1, 400, 1))
 temp_data = data[batch_idx,:]
 output = sess.run([y, q1, q2], feed_dict=x : temp_data )
 print(output)

The output of this is:

[0.5078107, [0.0, 0.0], [0.0, 0.0]]
[0.50185573, [0.0, 0.0], [0.0, 0.0]]
[0.4966678, [0.0, 0.0], [0.0, 0.0]]
[0.50407946, [0.0, 0.0], [0.0, 0.0]]
[0.49476147, [0.0, 0.0], [0.0, 0.0]]

I would appreciate any guidance provided....