tf_get_variable_example.py

# -*- coding: utf-8 -*-

# tf.get_variable & tf.variable_scope 예제 (변수가 공유된다.)

# MNIST 데이터를 다운로드 한다.
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

import tensorflow as tf

def softmax_classifier_vs(x, reuse_flag=False):	
	with tf.variable_scope("softmax_classifier", reuse=reuse_flag):
		W = tf.get_variable("W", [784, 10])
		b = tf.get_variable("b", [10])
		y = tf.nn.softmax(tf.matmul(x, W) + b)

		return y

x = tf.placeholder(tf.float32, [None, 784])
classifier1_vs = softmax_classifier_vs(x)
#classifier2_vs = softmax_classifier_vs(x_vs) # error!
classifier2_vs = softmax_classifier_vs(x, True)

# cross-entropy 모델을 설정한다.
y_ = tf.placeholder(tf.float32, [None, 10])
cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(classifier1_vs), reduction_indices=[1]))
train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)

# 경사하강법으로 모델을 학습한다.
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
for i in range(1000):
  batch_xs, batch_ys = mnist.train.next_batch(100)
  sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})

# 학습된 모델이 얼마나 정확한지를 출력한다.
# 변수가 공유되므로 classifier2_vs는 명시적으로 학습시키지 않았지만 classifier1_vs와 동일한 정확도를 갖는다.
correct_prediction1 = tf.equal(tf.argmax(classifier1_vs,1), tf.argmax(y_,1))
correct_prediction2 = tf.equal(tf.argmax(classifier2_vs,1), tf.argmax(y_,1))
accuracy1 = tf.reduce_mean(tf.cast(correct_prediction1, tf.float32))
accuracy2 = tf.reduce_mean(tf.cast(correct_prediction2, tf.float32))
print(sess.run(accuracy1, feed_dict={x: mnist.test.images, y_: mnist.test.labels}))
print(sess.run(accuracy2, feed_dict={x: mnist.test.images, y_: mnist.test.labels}))