21.txt

This complete matlab for neural network

发信人: jeff814 (mimi), 信区: DataMining
标  题: osu svm中的Parameters说明
发信站: 南京大学小百合站 (Fri Jun 13 16:00:02 2003)

整理了一下，供使用osu svm的朋友参考

参数Parameters：

|Kernel Type| Degree | Gamma | Coefficient | C |Cache size|epsilon| SVM Type |
	nu (nu-svm) |loss toleration | shrinking |		

Kernel Type:
 0 --- Linear   1 --- Polynomial   2 --- RBF    3 --- Sigmoid

degree: parameter needed for kernel of type polynomial (default:3)

Gamma: parameter needed for all kernels except linear. If the input value is z
ero, Gamma will be set defautly as 1/(max_pattern_dimension) in the function. 
If the input value is non-zero, Gamma will remain unchanged in the function.(d
efault: 1)

Coefficient: parameter needed for kernels of type polynomial and sigmoid (defa
ult:0)

C（线性不可分情况下对错分样本的惩罚系数）: Cost of the constrain violation (fo
r C-SVC & C-SVR) (default 1), or the nu for the nu-svm and 1-svm in the second
 stage. This nu should smaller than the nu defined in Parameter(9)

Cache Size: as the buffer to hold the <X(:,i),X(:,j)> (in MB)

epsilon（松弛项）: tolerance of termination criterion

SVM Type: (Recommend only using 0, which is c-SVM classifier)(default: 0)
                   0 --- c-SVM classifier
                   1 --- nu-SVM classifier
                   2 --- one-class SVM  （分布估计distribution estimation）
                   3 --- epsilon-SVR 
                   4 --- nu-SVR

nu: the nu of nu-SVM used in the boundary finding process。nu of nu-SVC, one-c
lass SVM, and nu-SVR (default: 0.5)

loss toleration: epsilon in loss function of epsilon-SVR (default: 0.1) 

shrinking: whether to use the shrinking heuristics, 0 or 1 (default: 1)

Percentage（SVMPlot2.m）: the percentage of input train samples used for bound
ary finding.	




--

※ 来源:．南京大学小百合站 http://bbs.nju.edu.cn [FROM: 202.99.41.202]