tensorflow之softmax

softmax 就是把值做个映射，映射到0-1之间，并且映射之后，和为1.

举个例子：

tt = tf.constant([1.0,2.0])
y = tf.nn.softmax(tt)

with tf.Session() as sess:
    print(sess.run(y))

输出

 注意看1

argmax：返回最大数的索引

labels1=[[0, 0, 1], [0, 1, 0] , [1, 0, 0]]
labels2 = tf.argmax(labels1,0 ) 
with tf.Session() as sess:
    l2 = sess.run( labels2 )
    print ('argmax output(sparse labels)', l2 ) 

返回：

 [0,0,1]的最大数的索引是2

 [0,1,0]的最大数索引是1

 [0,0,1]的最大数索引是0

cross_entropy = -tf.reduce_sum(y_*tf.log(y))

cross_entropy2=tf.reduce_sum(tf.nn.softmax_cross_entropy_with_logits(logits = logits,labels = y_))

这两个函数输出是一样的

实例如下：

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]])
#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(logits = logits,labels = y_))#dont forget tf.reduce_sum()!!
 
with tf.Session() as sess:
    softmax=sess.run(y)
    c_e = sess.run(cross_entropy)
    c_e2 = sess.run(cross_entropy2)
    print(logits)
    print("step1:softmax result=")
    print(softmax)
    print("step2:cross_entropy result=")
    print(c_e)
    print("Function(softmax_cross_entropy_with_logits) result=")
    print(c_e2)

 输出如下：