2.1TF模型持久化

Posted 2020-10-19 Mr.Zhao

目前tf只能保存模型中的variable变量，整个模型还不能保存，版本1.x

保存模型代码

import tensorflow as tf
import numpy as np

# Save to file
# remember to define the same dtype and shape when restore
v1 = tf.Variable(tf.constant(1.0,shape=[1]),  name=\'v1\')
v2 = tf.Variable(tf.constant(2.0,shape=[1]),  name=\'v2\')
result=v1+v2

# tf.initialize_all_variables() no long valid from
# 2017-03-02 if using tensorflow >= 0.12
if int((tf.__version__).split(\'.\')[1]) < 12 and int((tf.__version__).split(\'.\')[0]) < 1:
    init = tf.initialize_all_variables()
else:
    init = tf.global_variables_initializer()

saver = tf.train.Saver()

with tf.Session() as sess:
   sess.run(init)
   save_path = saver.save(sess,"save_model/save_pp.ckpt")
   print("Save to path: ", save_path)

文件结构如下

还原模型代码

################################################
# restore variables
# redefine the same shape and same type for your variables
v1 = tf.Variable(tf.constant(1.0,shape=[1]),  name=\'v1\')
v2 = tf.Variable(tf.constant(2.0,shape=[1]),  name=\'v2\')
result=v1+v2
# not need init step

saver = tf.train.Saver()
with tf.Session() as sess:
    saver.restore(sess, "./save_model/save_pp.ckpt")
    print("v:", sess.run(v1))
    print("result:", sess.run(result))

报错信息

未解决