转载softmax 归一函数

总结为：   将一组数变换为  总和为1,各个数为0~1之间的软性归一化结果。 

=========================================================

关于logistic regression中的softmax 函数

是用来柔化输出值，减小值之间的差。

用来归一化一组值到0~1之间，  总和为1. 

步骤为： 

1. 求出最大值max
2. 由exp表达式将各个值转化为0~1之间的数  x[i] =  exp(x[i] - max)
3. 求sum，归一化。 

[cpp]  view plain  copy

 print ?

1. void LogisticRegression::softmax(double *x) {                                     
2.                                                            
3.   double max = 0.0;                                                                
4.   double sum = 0.0;                                                                
5.                                                                                    
6.   for(int i=0; i<n_out; i++) if(max < x[i]) max = x[i];                            
7.   for(int i=0; i<n_out; i++) {                                                     
8.     x[i] = exp(x[i] - max);                                                        
9.     sum += x[i];                                                                   
10.   }                                                                                
11.                                                                                    
12.   for(int i=0; i<n_out; i++) x[i] /= sum;    
13. }