svm理论与实验之21: 自定义核函数的使用

徐海蛟博士


真实场景下，数据的特征可能比较复杂，系统提供的4种核函数或许达不到最佳效果，那么就需要自定义核函数了。当然，有很多大牛干这个事情，我们可以拿来使用，通过自定义核方式。


如何用？这时候不再把训练与测试数据文件作为输入参数了，而是使用核矩阵作为输入参数。


Assume there are L training instances x1, ..., xL . ... L行训练样本
Let K(x, y) be the kernel value of two instances x 与 y. The input formats are:
New training instance for xi:
<label> 0:i 1:K(xi,x1) ... L:K(xi,xL) 


New testing instance for any x:
<label> 0:? 1:K(x,x1) ... L:K(x,xL) 


That is, in the training file the first column must be the "ID" of xi. In testing, ? can be any value.


All kernel values including ZEROs must be explicitly provided. Any permutation or random subsets of the training/testing files are also valid (see examples below).


Note: the format is slightly different from the precomputed kernel
package released in libsvmtools earlier.


例子:
Assume the original training data has 3个four-feature instances, testing data has one instance:
15  1:1 2:1 3:1 4:1
45     2:3      4:3
25          3:1
-----------------------------------
15  1:1     3:1


若使用线性核, we have the following new training/testing sets:
15  0:1 1:4  2:6   3:1
45  0:2 1:6  2:18  3:0 
25  0:3 1:1  2:0   3:1
 -------------------------------------
15  0:? 1:2  2:0   3:1


? can be any value.


Any subset of the above training file is also valid. 例如,
25  0:3  1:1  2:0   3:1
45  0:2  1:6  2:18  3:0
意味着核矩阵是:
[K(2,2) K(2,3)] = [18 0]
[K(3,2) K(3,3)] = [0  1]