Question

有人可以解释我从这个简单的代码得到的奇怪结果吗？我做错了吗？为什么输入参数a_C和a_G会发生变化？与传递的值有什么不同？

#-----------------------------

def dummy_function(a_C, a_G):

  diff = tf.subtract(a_C, a_G)    
  sqr = tf.square(diff)

  return a_C, a_G, diff, sqr

#-----------------------------

tf.reset_default_graph()

with tf.Session() as test:

  tf.set_random_seed(1)

  a_C = tf.random_normal([1], mean=1, stddev=4)    
  a_G = tf.random_normal([1], mean=1, stddev=4)    
  a_C_returned, a_G_returned, diff, sqr = dummy_function(a_C, a_G)

  print("a_C = " + str(a_C.eval()))    
  print("a_G = " + str(a_G.eval()))    
  print("a_C_returned = " + str(a_C_returned.eval()))    
  print("a_G_returned = " + str(a_G_returned.eval()))    
  print("diff = " + str(diff.eval()))    
  print("sqr = " + str(sqr.eval()))

#-----------------------------    
# results

a_C = [-1.68344498]    
a_G = [-0.39043474]    
a_C_returned = [ 4.70364952]
a_G_returned = [ 0.84769011]
diff = [-9.30598831]
sqr = [ 25.68828583]

提前感谢您的帮助，最好的祝福， KASIA

Answer 1

你的a_C不是tf.random_normal的结果张量！这是在每个eval中获取随机数的操作。这是从不使用.eval()的最佳演示。

相反，您需要在一个运行中评估这些张量，如

import tensorflow as tf

def dummy_function(a_C, a_G):

  diff = tf.subtract(a_C, a_G)
  sqr = tf.square(diff)

  return a_C, a_G, diff, sqr

with tf.Session() as sess:

  tf.set_random_seed(1)

  a_C = tf.random_normal([1], mean=1, stddev=4)
  a_G = tf.random_normal([1], mean=1, stddev=4)
  a_C_returned, a_G_returned, diff, sqr = dummy_function(a_C, a_G)

  a_C_, a_G_, a_C_returned_, a_G_returned_, diff_, sqr_ = sess.run([a_C, a_G, a_C_returned, a_G_returned, diff, sqr])

  print("a_C = " + str(a_C_))
  print("a_G = " + str(a_G_))
  print("a_C_returned = " + str(a_C_returned_))
  print("a_G_returned = " + str(a_G_returned_))
  print("diff = " + str(diff_))
  print("sqr = " + str(sqr_))

这可以保证所有返回的结果都基于相同的条目节点（即a_C, a_g）

Tensorflow减去奇怪的结果

1 个答案: