Tensorflow 使用pb文件保存(恢复)模型计算图和参数实例详解
作者:brooknew 时间:2023-12-13 05:42:06
一、保存:
graph_util.convert_variables_to_constants 可以把当前session的计算图串行化成一个字节流(二进制),这个函数包含三个参数:参数1:当前活动的session,它含有各变量
参数2:GraphDef 对象,它描述了计算网络
参数3:Graph图中需要输出的节点的名称的列表
返回值:精简版的GraphDef 对象,包含了原始输入GraphDef和session的网络和变量信息,它的成员函数SerializeToString()可以把这些信息串行化为字节流,然后写入文件里:
constant_graph = graph_util.convert_variables_to_constants( sess, sess.graph_def , ['sum_operation'] )
with open( pbName, mode='wb') as f:
f.write(constant_graph.SerializeToString())
需要指出的是,如果原始张量(包含在参数1和参数2中的组成部分)不参与参数3指定的输出节点列表所指定的张量计算的话,这些张量将不会存在返回的GraphDef对象里,也不会被串行化写入pb文件。
二、恢复:
恢复时,创建一个GraphDef,然后从上述的文件里加载进来,接着输入到当前的session:
graph0 = tf.GraphDef()
with open( pbName, mode='rb') as f:
graph0.ParseFromString( f.read() )
tf.import_graph_def( graph0 , name = '' )
三、代码:
import tensorflow as tf
from tensorflow.python.framework import graph_util
pbName = 'graphA.pb'
def graphCreate() :
with tf.Session() as sess :
var1 = tf.placeholder ( tf.int32 , name='var1' )
var2 = tf.Variable( 20 , name='var2' )#实参name='var2'指定了操作名,该操作返回的张量名是在
#'var2'后面:0 ,即var2:0 是返回的张量名,也就是说变量
# var2的名称是'var2:0'
var3 = tf.Variable( 30 , name='var3' )
var4 = tf.Variable( 40 , name='var4' )
var4op = tf.assign( var4 , 1000 , name = 'var4op1' )
sum = tf.Variable( 4, name='sum' )
sum = tf.add ( var1 , var2, name = 'var1_var2' )
sum = tf.add( sum , var3 , name='sum_var3' )
sumOps = tf.add( sum , var4 , name='sum_operation' )
oper = tf.get_default_graph().get_operations()
with open( 'operation.csv','wt' ) as f:
s = 'name,type,output\n'
f.write( s )
for o in oper:
s = o.name
s += ','+ o.type
inp = o.inputs
oup = o.outputs
for iip in inp :
s #s += ','+ str(iip)
for iop in oup :
s += ',' + str(iop)
s += '\n'
f.write( s )
for var in tf.global_variables():
print('variable=> ' , var.name) #张量是tf.Variable/tf.Add之类操作的结果,
#张量的名字使用操作名加:0来表示
init = tf.global_variables_initializer()
sess.run( init )
sess.run( var4op )
print('sum_operation result is Tensor ' , sess.run( sumOps , feed_dict={var1:1}) )
constant_graph = graph_util.convert_variables_to_constants( sess, sess.graph_def , ['sum_operation'] )
with open( pbName, mode='wb') as f:
f.write(constant_graph.SerializeToString())
def graphGet() :
print("start get:" )
with tf.Graph().as_default():
graph0 = tf.GraphDef()
with open( pbName, mode='rb') as f:
graph0.ParseFromString( f.read() )
tf.import_graph_def( graph0 , name = '' )
with tf.Session() as sess :
init = tf.global_variables_initializer()
sess.run(init)
v1 = sess.graph.get_tensor_by_name('var1:0' )
v2 = sess.graph.get_tensor_by_name('var2:0' )
v3 = sess.graph.get_tensor_by_name('var3:0' )
v4 = sess.graph.get_tensor_by_name('var4:0' )
sumTensor = sess.graph.get_tensor_by_name("sum_operation:0")
print('sumTensor is : ' , sumTensor )
print( sess.run( sumTensor , feed_dict={v1:1} ) )
graphCreate()
graphGet()
四、保存pb函数代码里的操作名称/类型/返回的张量:
operation name | operation type | output | ||
var1 | Placeholder | Tensor("var1:0" | dtype=int32) | |
var2/initial_value | Const | Tensor("var2/initial_value:0" | shape=() | dtype=int32) |
var2 | VariableV2 | Tensor("var2:0" | shape=() | dtype=int32_ref) |
var2/Assign | Assign | Tensor("var2/Assign:0" | shape=() | dtype=int32_ref) |
var2/read | Identity | Tensor("var2/read:0" | shape=() | dtype=int32) |
var3/initial_value | Const | Tensor("var3/initial_value:0" | shape=() | dtype=int32) |
var3 | VariableV2 | Tensor("var3:0" | shape=() | dtype=int32_ref) |
var3/Assign | Assign | Tensor("var3/Assign:0" | shape=() | dtype=int32_ref) |
var3/read | Identity | Tensor("var3/read:0" | shape=() | dtype=int32) |
var4/initial_value | Const | Tensor("var4/initial_value:0" | shape=() | dtype=int32) |
var4 | VariableV2 | Tensor("var4:0" | shape=() | dtype=int32_ref) |
var4/Assign | Assign | Tensor("var4/Assign:0" | shape=() | dtype=int32_ref) |
var4/read | Identity | Tensor("var4/read:0" | shape=() | dtype=int32) |
var4op1/value | Const | Tensor("var4op1/value:0" | shape=() | dtype=int32) |
var4op1 | Assign | Tensor("var4op1:0" | shape=() | dtype=int32_ref) |
sum/initial_value | Const | Tensor("sum/initial_value:0" | shape=() | dtype=int32) |
sum | VariableV2 | Tensor("sum:0" | shape=() | dtype=int32_ref) |
sum/Assign | Assign | Tensor("sum/Assign:0" | shape=() | dtype=int32_ref) |
sum/read | Identity | Tensor("sum/read:0" | shape=() | dtype=int32) |
var1_var2 | Add | Tensor("var1_var2:0" | dtype=int32) | |
sum_var3 | Add | Tensor("sum_var3:0" | dtype=int32) | |
sum_operation | Add | Tensor("sum_operation:0" | dtype=int32) |
来源:https://blog.csdn.net/brooknew/article/details/83063512
标签:Tensorflow,pb,保存模型,计算图
![](/images/zang.png)
![](/images/jiucuo.png)
猜你喜欢
mysql远程跨库联合查询的示例
2024-01-13 23:09:10
![](https://img.aspxhome.com/file/2023/8/71098_0s.jpg)
如何在python中实现ECDSA你知道吗
2023-03-14 14:07:17
![](https://img.aspxhome.com/file/2023/5/110045_0s.jpg)
基于python+selenium自动健康打卡的实现代码
2022-04-28 06:31:28
![](https://img.aspxhome.com/file/2023/0/90780_0s.png)
MySQL中的主键自增机制详情
2024-01-18 23:50:54
Oracle生成单据编号存储过程的实例代码
2024-01-23 19:10:19
golang 调用c语言动态库方式实现
2024-05-22 17:51:02
Linux PHP连接MSSQL
2011-06-02 12:01:00
SQL Server中Update的用法
2008-12-29 13:57:00
Python中shutil模块的学习笔记教程
2022-12-22 20:42:05
Jupyter安装拓展nbextensions及解决官网下载慢的问题
2023-11-24 20:50:54
![](https://img.aspxhome.com/file/2023/8/133898_0s.jpg)
asp如何编写一个小数转换分数的函数?
2009-11-08 19:01:00
Elasticsearch的删除映射类型操作示例
2022-05-03 09:46:50
实现php删除链表中重复的结点
2023-09-05 09:36:15
VSCode配置python环境及中文问题解决方法
2022-07-14 15:39:02
![](https://img.aspxhome.com/file/2023/3/126923_0s.png)
python实现处理mysql结果输出方式
2024-01-28 03:30:42
Python接口测试数据库封装实现原理
2024-01-27 11:19:46
![](https://img.aspxhome.com/file/2023/5/117965_0s.png)
Python中logging日志库实例详解
2023-10-04 13:26:25
php计算函数执行时间的方法
2023-09-29 12:37:19
Python面向对象编程之类的继承
2021-10-27 23:46:39
![](https://img.aspxhome.com/file/2023/6/83896_0s.png)
python判断文件是否存在,不存在就创建一个的实例
2022-04-29 02:28:55