精选留言(8)

• 
  陈浩然

  2019-01-14

  import tensorflow as tf
  
  W = tf.Variable(initial_value = tf.random_normal(shape = (1 , 4) ,mean = 100 , stddev = 0.35 ), name = "W")
  b = tf.Variable(tf.zeros([4]) , name="b")
  [W , b]
  [<tf.Variable 'W:0' shape=(1, 4) dtype=float32_ref>,
   <tf.Variable 'b:0' shape=(4,) dtype=float32_ref>]
  
  sess = tf.Session()
  sess.run(tf.global_variables_initializer())
  sess.run( [W , b])
  [array([[100.40549 , 99.182625, 100.23513 , 99.42683 ]], dtype=float32),
   array([0., 0., 0., 0.], dtype=float32)]
  
  sess.run(tf.assign_add(b , [1,1,1,1] ))
  
  sess.run(tf.assign_add(b , [1,1,1,1] ))
  ​
  array([1., 1., 1., 1.], dtype=float32)
  
  sess.run(b)
  array([1., 1., 1., 1.], dtype=float32)
  
  saver = tf.train.Saver({'W' : W , 'b' : b})
  saver.save(sess , './summary/data.ckpt' , global_step = 0)
  WARNING:tensorflow:Issue encountered when serializing trainable_variables.
  Type is unsupported, or the types of the items don't match field type in CollectionDef. Note this is a warning and probably safe to ignore.
  'test' has type str, but expected one of: int, long, bool
  WARNING:tensorflow:Issue encountered when serializing variables.
  Type is unsupported, or the types of the items don't match field type in CollectionDef. Note this is a warning and probably safe to ignore.
  'test' has type str, but expected one of: int, long, bool
  './summary/data.ckpt-0'
  
  sess.run(tf.assign_add(b , [1,1,1,1] ))
  sess.run(b)
  array([2., 2., 2., 2.], dtype=float32)
  
  saver.restore(sess , './summary/data.ckpt-0')
  
  saver.restore(sess , './summary/data.ckpt-0')
  sess.run(b)
  INFO:tensorflow:Restoring parameters from ./summary/data.ckpt-0
  array([1., 1., 1., 1.], dtype=float32)

  展开

  

   2
• 
  bearlu

  2019-01-12

  老师，麻烦贴一下源码地址

  作者回复: https://github.com/DjangoPeng/tensorflow-101

  

   2
• 
  ZeroIce

  2019-08-24

  老师，有什么书籍推荐看的？英文系列也行

  作者回复: 《Python深度学习》Keras作者写的
  《深入理解TensorFlow》也可以看看

  

  
• 
  pyw

  2019-08-08

  从文件恢复的时候出错：
  ```
  WARNING: Logging before flag parsing goes to stderr.
  W0808 00:16:28.798274 140416759756608 deprecation.py:323] From /home/pyw/.local/lib/python3.6/site-packages/tensorflow/python/training/saver.py:1276: checkpoint_exists (from tensorflow.python.training.checkpoint_management) is deprecated and will be removed in a future version.
  Instructions for updating:
  Use standard file APIs to check for files with this prefix.
  Traceback (most recent call last):
    File "c20.py", line 39, in <module>
      saver.restore(sess, './save/1.save')
    File "/home/pyw/.local/lib/python3.6/site-packages/tensorflow/python/training/saver.py", line 1278, in restore
      compat.as_text(save_path))
  ValueError: The passed save_path is not a valid checkpoint: ./save/1.save
  ```
  
  c20.py的39行是：
  saver.restore(sess, './save/1.save')
  

  展开

  

  
• 
  pyw

  2019-08-07

  有的方法过期了
  ```
  W0807 23:51:48.566468 140651389855552 deprecation_wrapper.py:119] From c20.py:14: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.
  W0807 23:51:48.864450 140651389855552 deprecation_wrapper.py:119] From c20.py:15: The name tf.global_variables_initializer is deprecated. Please use tf.compat.v1.global_variables_initializer instead.
  W0807 23:51:49.436012 140651389855552 deprecation_wrapper.py:119] From c20.py:19: The name tf.assign_add is deprecated. Please use tf.compat.v1.assign_add instead.
  ```
  
  如果按照提示修改，会得到
  
  ```
      sess.run(tf.compat.v1.global_variables_initializer())
  TypeError: unbound method run() must be called with Session instance as first argument (got Operation instance instead)
  ```

  展开

   1

  
• 
  沉枫y

  2019-02-26

  接上一条留言
  print(sess.run(b))
  print(b) 两者的输出结果不一样
  [1. 1. 1. 1.]
  <tf.Variable 'b:0' shape=(4,) dtype=float32_ref>
  是不是可以理解为 b的本身只是一个定义好的30位的浮点数张量
  在经过session会话后，只是在该会话改变了他的值，但并不改变他本身的属性
  还是以上理解是错的 是其他的解释

  展开

  作者回复: 在该会话（上下文）中改变（初始化、计算、赋值）了它。如果你关闭会话 ，它还是一个数据流图中的Variable 实例。

  

  
• 
  沉枫y

  2019-02-26

  执行完sess.run([W, b])后 去打印变量W后的结果并没有发生变化（在IDLE中操作）

  作者回复: print(sess.run([W, b])) 会打印出结果，因为会话执行特定节点（变量、操作）返回的是 python 数据类型。但是 W和b本身是 TensorFlow DSL 定义的数据类型，是一个抽象的“壳”。无论你执行多少次，“壳”是不会变得。

  

  
• 
  稀饭ETF

  2019-01-13

  终于赶上进度了😂

  作者回复: 加油加油!

  

  