tf.control_dependencies()
2018-04-02 19:36
537 查看
tf.control_dependencies()设计是用来控制计算流图的,给图中的某些计算指定顺序。比如:我们想要获取参数更新后的值,那么我们可以这么组织我们的代码。
opt = tf.train.Optimizer().minize(loss) with tf.control_dependencies([opt]): updated_weight = tf.identity(weight) with tf.Session() as sess: tf.global_variables_initializer().run() sess.run(updated_weight, feed_dict={...}) # 这样每次得到的都是更新后的weight关于tf.control_dependencies的具体用法,請移步官网https://www.tensorflow.org/api_docs/python/tf/Graph#control_dependencies,总结一句话就是,在执行某些
op,tensor之前,某些
op,tensor得首先被运行。
下面说明两种 control_dependencies 不 work 的情况
下面有两种情况,control_dependencies不work,其实并不是它真的不work,而是我们的使用方法有问题。第一种情况:
import tensorflow as tf w = tf.Variable(1.0) ema = tf.train.ExponentialMovingAverage(0.9) update = tf.assign_add(w, 1.0) ema_op = ema.apply([update]) with tf.control_dependencies([ema_op]): ema_val = ema.average(update) with tf.Session() as sess: tf.global_variables_initializer().run() for i in range(3): print(sess.run([ema_val]))也许你会觉得,在我们
sess.run([ema_val]),
ema_op都会被先执行,然后再计算
ema_val,实际情况并不是这样,为什么?
有兴趣的可以看一下源码,就会发现
ema.average(update)不是一个
op,它只是从
ema对象的一个字典中取出键对应的
tensor而已,然后赋值给
ema_val。这个
tensor是由一个在
tf.control_dependencies([ema_op])外部的一个
op计算得来的,所以
control_dependencies会失效。解决方法也很简单,看代码:
import tensorflow as tf w = tf.Variable(1.0) ema = tf.train.ExponentialMovingAverage(0.9) update = tf.assign_add(w, 1.0) ema_op = ema.apply([update]) with tf.control_dependencies([ema_op]): ema_val = tf.identity(ema.average(update)) #一个identity搞定 with tf.Session() as sess: tf.global_variables_initializer().run() for i in range(3): print(sess.run([ema_val]))第二种情况: 这个情况一般不会碰到,这是我在测试
control_dependencies发现的
import tensorflow as tf w = tf.Variable(1.0) ema = tf.train.ExponentialMovingAverage(0.9) update = tf.assign_add(w, 1.0) ema_op = ema.apply([update]) with tf.control_dependencies([ema_op]): w1 = tf.Variable(2.0) ema_val = ema.average(update) with tf.Session() as sess: tf.global_variables_initializer().run() for i in range(3): print(sess.run([ema_val, w1]))这种情况下,
control_dependencies也不 work。读取
w1的值并不会触发
ema_op, 原因请看代码:
#这段代码出现在Variable类定义文件中第287行, # 在创建Varible时,tensorflow是移除了dependencies了的 #所以会出现 control 不住的情况 with ops.control_dependencies(None): ... ###################################################################################################参考这里点击打开链接的信息我们可以知道,TF可以协调多个数据流,在存在依赖的节点下非常有用,例如节点B要读取模型参数值V更新后的值,而节点A负责更新参数V,所以节点B就要等节点A执行完成后再执行,不然读到的就是更新以前的数据。这时候就需要个运算控制器tf.control_dependencies。
参考官方说明文档
format:control_dependencies(self, control_inputs)
arguments:control_inputs: A list of `Operation` or `Tensor` objects which must be executed or computed before running the operations defined in the context. (注意这里control_inputs是list)
return: A context manager that specifies control dependencies for all operations constructed within the context.(返回所有在环境中的控制依赖的上下文管理器)
其实用法很简单,只有在 control_inputs被执行以后,上下文管理器中的操作才会被执行。例如
[python] view plain copy with tf.control_dependencies([a, b, c]):
# `d` and `e` will only run after `a`, `b`, and `c` have executed.
d = ...
e = ...
只有[a,b,c]都被执行了才会执行d和e操作,这样就实现了流的控制。当然,官方文档里还介绍了嵌套多个流控制
[python] view plain copy with tf.control_dependencies([a, b]):
# Ops constructed here run after `a` and `b`.
with tf.control_dependencies([c, d]):
# Ops constructed here run after `a`, `b`, `c`, and `d`
也能通过参数None清除控制依赖例如
[python] view plain copy with g.control_dependencies([a, b]):
# Ops constructed here run after `a` and `b`.
with g.control_dependencies(None):
# Ops constructed here run normally, not waiting for either `a` or `b`.
with g.control_dependencies([c, d]):
# Ops constructed here run after `c` and `d`, also not waiting
# for either `a` or `b`.
相关文章推荐
- tensorflow API:tf.control_dependencies
- (Tensorflow之十)tf.control_dependencies()用法
- tf.identity 和 tf.control_dependencies的用法
- tf.control_dependencies()和tf.identity()
- TensorFlow入门基础知识(七)tf.control_dependencies的用法
- tf.control_dependencies与tf.identity组合详解
- tf.control_dependencies()作用及用法
- tf.control_dependencies()作用及用法
- tf.control_dependencies()
- Tensorflow 解惑[一]:关于tf.identity 和tf.control_dependencies
- Could not load file or assembly ‘AjaxControlToolkit’ or one of its dependencies. The parameter is incorrect.
- Could not load file or assembly ‘AjaxControlToolkit’ or one of its dependencies. The parameter is incorrect. (Exception
- TensorFlow流程控制函数:tf.group tf.control_dependencies
- CruiseControl Enterprise 最佳实践 (2) : Keep your dependencies to yourself
- 用C#实现MVC(Model View Control)模式介绍
- Maven instal Could not resolve dependencies for project 找不到本地仓库下的jar
- managing control files
- No 'Access-Control-Allow-Origin' header is present on the requested resource解决方法
- 使用control.invoke返回当前窗口线程[转]
- 如何使用VC中的Custom Control?