Saver与Restore

...
# Create a saver.
saver = tf.train.Saver(...variables...)
# Launch the graph and train, saving the model every 1,000 steps.
sess = tf.Session()
for step in xrange(1000000):
    sess.run(..training_op..)
    if step % 1000 == 0:
        # Append the step number to the checkpoint name:
        saver.save(sess, ‘my-model‘, global_step=step)

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)

import tensorflow as tf
import time

time.clock()

x = tf.placeholder(tf.float32 ,[None, 784])
W = tf.Variable(tf.zeros([784,10]))
b = tf.Variable(tf.zeros([10]))
y = tf.nn.softmax(tf.matmul(x,W) + b)

# 为了计算交叉熵，我们需要添加一个新的占位符用于输入正确值。
y_ = tf.placeholder(tf.float32, [None,10])
cross_entropy = -tf.reduce_sum(y_*tf.log(y))
# 在此，我们要求TF使用梯度下降算法，并以0.01的学习速率最小化交叉熵。
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy)

init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)

# 创建Saver节点，并设置自动保存最近n=1次模型
saver = tf.train.Saver(max_to_keep=1)
saver_max_acc = 0 
for i in range(100):
    batch_xs, batch_ys = mnist.train.next_batch(100)
    sess.run(train_step, feed_dict={x: batch_xs, y_:batch_ys})
    correct_prediction = tf.equal(tf.argmax(y,1), tf.arg_max(y_, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction,"float"))
    if (i+1)%10 == 0:
        print(‘{0:0>2d}:{1:.4f}‘.format((i+1),accuracy.eval(session=sess, feed_dict={x: mnist.test.images, y_: mnist.test.labels})))
    # 添加判断语句，选择保存精度最高的模型
    if accuracy > saver_max_acc:
        saver.save(sess,‘ckpt/mnist.ckpt‘,global_step=i+1)
        saver_max_acc = accuracy
sess.close()
print(time.clock())

restore(sess, save_path)
# sess: A Session to use to restore the parameters.
# save_path: Path where parameters were previously saved.