关于tf.nn.softmax_cross_entropy_with_logits的使用

第一个参数不用管。输入labels时，必须为one-hot形式。

若不是one-hot形式，只用每一次对应的真值，则会自动广播补全，实验如下：

import tensorflow as tf  
import numpy as np 

#our NN's output  
logits=tf.constant([[1.0,2.0,3.0],[1.0,2.0,3.0],[1.0,2.0,3.0]])  
#step1:do softmax  
y=tf.nn.softmax(logits)  
#true label  
y_=tf.constant([[0.0,0.0,1.0],[0.0,0.0,1.0],[0.0,0.0,1.0]])  

y_2 = np.zeros(3)
for i in range(3):
    y_2[i] = 1

y_3 = np.zeros(3)
y_3[2] = 1


#step2:do cross_entropy  
cross_entropy = -tf.reduce_sum(y_*tf.log(y))  
#do cross_entropy just one step  
cross_entropy2=tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits(labels=y_2, logits=logits))#dont forget tf.reduce_sum()!!

cross_entropy3=tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits(labels=y_3, logits=logits))#dont forget tf.reduce_sum()!!
  
with tf.Session() as sess:   
    c_e = sess.run(cross_entropy)  
    c_e2 = sess.run(cross_entropy2)
    c_e3 = sess.run(cross_entropy3)  
    print("step1 : cross_entropy result=")  
    print(c_e)  
    print("step2 : result=")  
    print(c_e2)
    print("step3 : result=")  
    print(c_e3) 

结果如下：

step1 : cross_entropy result=
1.222818
step2 : result=
12.668453
step3 : result=
1.2228179