脚本之家,脚本语言编程技术及教程分享平台!
分类导航

Python|VBS|Ruby|Lua|perl|VBA|Golang|PowerShell|Erlang|autoit|Dos|bat|

服务器之家 - 脚本之家 - Python - tensorflow模型保存、加载之变量重命名实例

tensorflow模型保存、加载之变量重命名实例

2020-04-05 12:26* star * Python

今天小编就为大家分享一篇tensorflow模型保存、加载之变量重命名实例,具有很好的参考价值,希望对大家有所帮助。一起跟随小编过来看看吧

话不多说,干就完了。

变量重命名的用处?

简单定义:简单来说就是将模型A中的参数parameter_A赋给模型B中的parameter_B

使用场景:当需要使用已经训练好的模型参数,尤其是使用别人训练好的模型参数时,往往别人模型中的参数命名方式与自己当前的命名方式不同,所以在加载模型参数时需要对参数进行重命名,使得代码更简洁易懂。

实现方法:

1)、模型保存

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
import os
import tensorflow as tf
 
weights = tf.Variable(initial_value=tf.truncated_normal(shape=[1024, 2],
                            mean=0.0,
                            stddev=0.1),
           dtype=tf.float32,
           name="weights")
biases = tf.Variable(initial_value=tf.zeros(shape=[2]),
           dtype=tf.float32,
           name="biases")
 
weights_2 = tf.Variable(initial_value=weights.initialized_value(),
            dtype=tf.float32,
            name="weights_2")
 
# saver checkpoint
if os.path.exists("checkpoints") is False:
  os.makedirs("checkpoints")
 
saver = tf.train.Saver()
with tf.Session() as sess:
  init_op = [tf.global_variables_initializer()]
  sess.run(init_op)
  saver.save(sess=sess, save_path="checkpoints/variable.ckpt")

2)、模型加载(变量名称保持不变)

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
import tensorflow as tf
from matplotlib import pyplot as plt
import os
 
current_path = os.path.dirname(os.path.abspath(__file__))
 
def restore_variable(sess):
  # need not initilize variable, but need to define the same variable like checkpoint
  weights = tf.Variable(initial_value=tf.truncated_normal(shape=[1024, 2],
                              mean=0.0,
                              stddev=0.1),
             dtype=tf.float32,
             name="weights")
  biases = tf.Variable(initial_value=tf.zeros(shape=[2]),
             dtype=tf.float32,
             name="biases")
 
  weights_2 = tf.Variable(initial_value=weights.initialized_value(),
              dtype=tf.float32,
              name="weights_2")
 
  saver = tf.train.Saver()
 
  ckpt_path = os.path.join(current_path, "checkpoints", "variable.ckpt")
  saver.restore(sess=sess, save_path=ckpt_path)
 
  weights_val, weights_2_val = sess.run(
    [
      tf.reshape(weights, shape=[2048]),
      tf.reshape(weights_2, shape=[2048])
    ]
  )
 
  plt.subplot(1, 2, 1)
  plt.scatter([i for i in range(len(weights_val))], weights_val)
  plt.subplot(1, 2, 2)
  plt.scatter([i for i in range(len(weights_2_val))], weights_2_val)
  plt.show()
 
 
if __name__ == '__main__':
  with tf.Session() as sess:
    restore_variable(sess)

3)、模型加载(变量重命名)

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
import tensorflow as tf
from matplotlib import pyplot as plt
import os
 
current_path = os.path.dirname(os.path.abspath(__file__))
 
 
def restore_variable_renamed(sess):
  conv1_w = tf.Variable(initial_value=tf.truncated_normal(shape=[1024, 2],
                              mean=0.0,
                              stddev=0.1),
             dtype=tf.float32,
             name="conv1_w")
  conv1_b = tf.Variable(initial_value=tf.zeros(shape=[2]),
             dtype=tf.float32,
             name="conv1_b")
 
  conv2_w = tf.Variable(initial_value=conv1_w.initialized_value(),
             dtype=tf.float32,
             name="conv2_w")
 
  # variable named 'weights' in ckpt assigned to current variable conv1_w
  # variable named 'biases' in ckpt assigned to current variable conv1_b
  # variable named 'weights_2' in ckpt assigned to current variable conv2_w
  saver = tf.train.Saver({
    "weights": conv1_w,
    "biases": conv1_b,
    "weights_2": conv2_w
  })
 
  ckpt_path = os.path.join(current_path, "checkpoints", "variable.ckpt")
  saver.restore(sess=sess, save_path=ckpt_path)
 
  conv1_w__val, conv2_w__val = sess.run(
    [
      tf.reshape(conv1_w, shape=[2048]),
      tf.reshape(conv2_w, shape=[2048])
    ]
  )
 
  plt.subplot(1, 2, 1)
  plt.scatter([i for i in range(len(conv1_w__val))], conv1_w__val)
  plt.subplot(1, 2, 2)
  plt.scatter([i for i in range(len(conv2_w__val))], conv2_w__val)
  plt.show()
 
 
if __name__ == '__main__':
  with tf.Session() as sess:
    restore_variable_renamed(sess)

总结:

# 之前模型中叫 'weights'的变量赋值给当前的conv1_w变量

# 之前模型中叫 'biases' 的变量赋值给当前的conv1_b变量

# 之前模型中叫 'weights_2'的变量赋值给当前的conv2_w变量

saver = tf.train.Saver({

"weights": conv1_w,

"biases": conv1_b,

"weights_2": conv2_w

})

以上这篇tensorflow模型保存、加载之变量重命名实例就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。

原文链接:https://blog.csdn.net/cxx654/article/details/88927962

延伸 · 阅读

精彩推荐