可能会随时更新,,,
第一个实验(desktop computer)
将2224个ep和normal总样本进行归一化:
D:\Program Files\MATLAB\R2013a\toolbox\libsvm-3.20\windows>svm-scale.exe epnorma
llibsvm.svm > epnormallibsvm.scale
WARNING: original #nonzeros 2127904
new #nonzeros 6671908
Use -l 0 if many original feature values are zeros
对2224个总样本抽样(按约8:2的比例分为训练集和测试集):
D:\Program Files\MATLAB\R2013a\toolbox\libsvm-3.20\tools>python subset.py epnorm
allibsvm.scale 1770 epnormallibsvm.train epnormallibsvm.test
使用最easy的方式跑第一次svm,即使用easy.py工具(需要安装gnuplot软件):
D:\Program Files\MATLAB\R2013a\toolbox\libsvm-3.20\tools>python easy.py epnormal
libsvm.train epnormallibsvm.test
Scaling training data...
WARNING: original #nonzeros 5309923
new #nonzeros 5309939
Use -l 0 if many original feature values are zeros
Cross validation...
艾玛,机子卡了,,,
the second day...
D:\Program Files\MATLAB\R2013a\toolbox\libsvm-3.20\windows>svm-predict.exe epnor
mallibsvm.test.scale epnormallibsvm.train.model epnormalresult1.txt
Accuracy = 97.7974% (444/454) (classification)
结果还不错啊。