Saver类--变量的保存和恢复

保存和恢复变量

我们经常在训练完一个模型之后希望保存训练的结果，这些结果指的是模型的参数，以便下次迭代的训练或者用作测试。Tensorflow针对这一需求提供了Saver类。

1. Saver类提供了向checkpoints文件保存和从checkpoints文件中恢复变量的相关方法。Checkpoints文件是一个二进制文件，它把变量名映射到对应的tensor值 。
2. 只要提供一个计数器，当计数器触发时，Saver类可以自动的生成checkpoint文件。这让我们可以在训练过程中保存多个中间结果。例如，我们可以保存每一步训练的结果。
3. 为了避免填满整个磁盘，Saver可以自动的管理Checkpoints文件。例如，我们可以指定保存最近的N个Checkpoints文件。

saver.save(sess, 'my-model', global_step=0) ==> filename: 'my-model-0'
saver.save(sess, 'my-model', global_step=1000) ==> filename: 'my-model-1000'

保存方法

import tensorflow as tf 
# Create some variables.
v1 = tf.Variable(tf.ones(shape=[2,3]), name="v1")
v2 = tf.Variable(tf.ones(shape=[2,2]), name="v2")
# Add an op to initialize the variables.
init_op = tf.global_variables_initializer()
# Add ops to save and restore all the variables.
saver = tf.train.Saver()
# Later, launch the model, initialize the variables, do some work, save the
# variables to disk.
with tf.Session() as sess:
  sess.run(init_op)
  # Do some work with the model.
  # Save the variables to disk.
  save_path = saver.save(sess, "./tmp/model.ckpt",global_step=100)
  print("Model saved in file:%s "%(save_path))

恢复变量用tf.train.Saver().restore(),例子，恢复的变量，不需要事先初始化

# Create some variables.
import tensorflow as tf
v1 = tf.Variable(tf.ones(shape=[2,3]), name="v1")
v2 = tf.Variable(tf.ones(shape=[2,2]), name="v2")
# Add ops to save and restore all the variables.
saver = tf.train.Saver()
# Later, launch the model, use the saver to restore variables from disk, and
# do some work with the model.
with tf.Session() as sess:
  # Restore variables from disk.
    saver.restore(sess, "./tmp/model.ckpt-100")
    print("v1 name is%s v1=%s"%(v1.name,sess.run(v1)))
    print("v2 name is%s v1=%s"%(v2.name,sess.run(v2)))
  # Do some work with the model

选择存储和恢复哪些变量

（这里先大概写一些简单的东西，以后碰到更高级的应用再回来补充，历史总是螺旋形的上升嘛，学习也一样，不能指望一口吃成胖子）

先指定保存变量V2到 ./board/model.ckpt的检查点文件中

import tensorflow as tf 
# Create some variables.
v1 = tf.Variable(tf.ones(shape=[2,3]), name="v1")
v2 = tf.Variable(tf.ones(shape=[2,2]), name="v2")
# Add an op to initialize the variables.
init_op = tf.global_variables_initializer()
# Add ops to save and restore all the variables.
saver = tf.train.Saver({'my_v2':v2})
# Later, launch the model, initialize the variables, do some work, save the
# variables to disk.
with tf.Session() as sess:
    sess.run(init_op)
    # Do some work with the model.
    # Save the variables to disk.
    save_path = saver.save(sess, "./board/model.ckpt",global_step=100)
    print("Model saved in file:%s "%(save_path))

再从检查点文件model-100中恢复V2出来

import tensorflow as tf 
# Create some variables.

v1 = tf.Variable(tf.zeros(shape=[2,3]), name="v1")
v2 = tf.Variable(tf.zeros(shape=[2,2]), name="v2")
# Add ops to save and restore all the variables.
saver = tf.train.Saver({'my_v2':v2})
# Later, launch the model, use the saver to restore variables from disk, and
# do some work with the model.
with tf.Session() as sess:
  # Restore variables from disk.
    sess.run(tf.global_variables_initializer())
    saver.restore(sess, "./board/model.ckpt-100")#注意，这时就不用再对变量用进行初始化了
    print("v2 name is%s v2=%s"%(v2.name,sess.run(v2)))

    print("v1 name is%s v1=%s"%(v1.name,sess.run(v1)))

尝试了一下，先进行初始化，再恢复变量，好像对结果并没有什么影响，可能初始化V1，V2变量之后，saver.restore（）又将V2变量初始了了一次吧。