基于网格模式搜索的支持向量机模型选择  

Gird-pattern method for model selection of support vector machines

作　　者：李兵[1] 姚全珠[1] 罗作民[1] 田元[1] 王伟[1] 

机构地区：[1]西安理工大学计算机科学与工程学院,西安710048

出　　处：《计算机工程与应用》2008年第15期136-138,共3页Computer Engineering and Applications

摘　　要：支持向量机的模型选择问题就是对于一个给定的核函数,调节核参数和惩罚因子C。分析了网格搜索算法和模式搜索算法,通过结合上述两种算法的优点提出了网格模式搜索算法。其核心原理是先用网格算法在全局范围内进行快速搜索,找到最优解的最小区间,再在这个最小区间内用模式搜索算法找到最优解。实验证明,网格模式搜索具有学习精度高和速度快的优点。For fixed functional form of the kernel,model selection amounts to tuning kernel parameters and the slack penalty coefficient C.Based on an analysis of the grid algorithm and pattern algorithm,this paper proposes a grid-pattern search algorithm,which combines grid search and pattern search.The main procedure of the proposed method include a fast search in the global domain with grid algorithm,then after obtaining the least interval containing the optimal solution,a pattern algorithm is employed to get the optimal solution in the interval.Experimental results indicate that this method has the advantage of high accuracy and speed when training SVM.

关 键 词：支持向量机 模型选择 网格模式搜索 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]