tensorflow - 如何将变量复制到 tensorflow 中的另一个图

Question

我希望将 tensorflow 变量从旧图复制到新图，然后删除旧图并将新图设置为默认图。下面是我的代码，但它引发了一个 AttributeError: 'Graph' object has no attribute 'variable1'。我是 tensorflow 新手。谁能给我一个具体的例子吗？

import tensorflow as tf
import numpy as np

graph1 = tf.Graph()
with graph1.as_default():
    variable1 = tf.Variable(np.array([2.,-1.]), name='x1')
    initialize = tf.initialize_all_variables()

sess1 = tf.Session(graph=graph1)
sess1.run(initialize)
print('sess1:',variable1.eval(session=sess1))

graph2 = tf.Graph()
with graph2.as_default():
    variable2 = tf.contrib.copy_graph.copy_variable_to_graph(graph1.variable1, graph2)
sess2 = tf.Session(graph=graph2)
#I want to remove graph1 and sess1, and make graph2 and sess2 the default here.
print('sess2:', variable2.eval(session=sess2))

Answer 1

1. tf.initialize_all_variables() 已弃用。请改用 tf.global_variables_initializer()。

2. 您不需要graph1.variable1。只需传递 variable1。

3. 您忘记在第二个 session 中初始化变量:

   initialize2 = tf.global_variables_initializer()
   sess2=tf.Session(graph=graph2)
   sess2.run(initialize2)
   

所以你的代码应该是这样的:

import tensorflow as tf
import numpy as np

graph1 = tf.Graph()
with graph1.as_default():
    variable1 = tf.Variable(np.array([2.,-1.]), name='x1')
    initialize = tf.global_variables_initializer()
    sess1=tf.Session(graph=graph1)
    sess1.run(initialize)
    print('sess1:',variable1.eval(session=sess1))

graph2 = tf.Graph()
with graph2.as_default():
    variable2=tf.contrib.copy_graph.copy_variable_to_graph(variable1,graph2)
    initialize2 = tf.global_variables_initializer()
    sess2=tf.Session(graph=graph2)
    sess2.run(initialize2)
    #I want to remove graph1 and sess1, ande make graph2 and sess2 as default here.
    print('sess2:',variable2.eval(session=sess2))