import tensorflow as tf
a_1 = tf.Variable(1)
b_1 = tf.Variable(2)
update_op1 = tf.assign(a_1, 10)
add = tf.add(a_1, b_1)
a_2 = tf.Variable(1)
b_2 = tf.Variable(2)
update_op2 = tf.assign(a_2, 10)
with tf.control_dependencies([update_op1,update_op2]):#tf.control_dependencies,该函数保证其辖域中的操作必须要在该函数所传递的参数中的操作完成后再进行
add_with_dependencies = tf.add(a_2, a_1)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
ans_1, ans_2 = sess.run([add, add_with_dependencies])
print("Add: ", ans_1)
print("Add_with_dependency: ", ans_2)
输出:
Add: 12
Add_with_dependency: 20
import tensorflow as tf
a_1 = tf.Variable(1)
b_1 = tf.Variable(2)
update_op = tf.assign(a_1, 10)
add = tf.add(a_1, b_1)
a_2 = tf.Variable(1)
b_2 = tf.Variable(2)
update_op = tf.assign(a_2, 10)
with tf.control_dependencies([update_op]):
add_with_dependencies = tf.add(a_2, b_2)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
ans_1, ans_2 = sess.run([add, add_with_dependencies])
print("Add: ", ans_1)
print("Add_with_dependency: ", ans_2)
输出:
Add: 3
Add_with_dependency: 12
本文通过两个示例详细解析了TensorFlow中control_dependencies函数的作用,展示了如何确保在执行某些操作前,先完成特定变量的更新。这对于理解并正确使用TensorFlow的计算图控制流至关重要。
1149

被折叠的 条评论
为什么被折叠?



