1、Good is good, but better carries it.精益求精，善益求善。MatlabSVM工具箱快速入手簡易教程-SVM工具箱快速入手簡易教程(byfaruto)一.matlab自帶的函數(shù)(matlab幫助文件里的例子)只有較新版本的matlab中有這兩個SVM的函數(shù)=svmtrainsvmclassify=簡要語法規(guī)則=svmtrainTrainsupportvectormachineclassifierSyntaxSVMStruct=svmtrain(Training,Group)SVMStruct=svmtrain(.,Kernel_Function,Kernel_F

2、unctionValue,.)SVMStruct=svmtrain(.,RBF_Sigma,RBFSigmaValue,.)SVMStruct=svmtrain(.,Polyorder,PolyorderValue,.)SVMStruct=svmtrain(.,Mlp_Params,Mlp_ParamsValue,.)SVMStruct=svmtrain(.,Method,MethodValue,.)SVMStruct=svmtrain(.,QuadProg_Opts,QuadProg_OptsValue,.)SVMStruct=svmtrain(.,SMO_Opts,SMO_OptsValu

3、e,.)SVMStruct=svmtrain(.,BoxConstraint,BoxConstraintValue,.)SVMStruct=svmtrain(.,Autoscale,AutoscaleValue,.)SVMStruct=svmtrain(.,Showplot,ShowplotValue,.)-svmclassifyClassifydatausingsupportvectormachineSyntaxGroup=svmclassify(SVMStruct,Sample)Group=svmclassify(SVMStruct,Sample,Showplot,ShowplotValu

4、e)=實例研究=loadfisheriris%載入matlab自帶的數(shù)據(jù)有關數(shù)據(jù)的信息可以自己到UCI查找,這是UCI的經(jīng)典數(shù)據(jù)之一,得到的數(shù)據(jù)如下圖:tu1HYPERLINK/attachment.php?aid=24862&k=5efc27ffe8e1ec9b74c29268689e3889&t=1263976450¬humb=yest_blank1.jpg(7.94KB)2009-5-1219:50其中meas是150*4的矩陣代表著有150個樣本每個樣本有4個屬性描述,species代表著這150個樣本的分類.data=meas(:,1),meas(:,2);%在這里只取meas

5、的第一列和第二列,即只選取前兩個屬性.groups=ismember(species,setosa);%由于species分類中是有三個分類:setosa,versicolor,virginica,為了使問題簡單,我們將其變?yōu)槎诸悊栴}:Setosaandnon-Setosa.train,test=crossvalind(holdOut,groups);cp=classperf(groups);%隨機選擇訓練集合測試集有關crossvalind的使用請自己help一下其中cp作用是后來用來評價分類器的.svmStruct=svmtrain(data(train,:),groups(train)

6、,showplot,true);%使用svmtrain進行訓練,得到訓練后的結構svmStruct,在預測時使用.訓練結果如圖:tu2HYPERLINK/attachment.php?aid=24863&k=d5871b5f3333ec8c873d178965ef6484&t=1263976450¬humb=yest_blank2.jpg(26.86KB)2009-5-1219:50classes=svmclassify(svmStruct,data(test,:),showplot,true);%對于未知的測試集進行分類預測,結果如圖:tu3HYPERLINK/attachment.p

7、hp?aid=24864&k=79da636571ae939195a632fa21c7c370&t=1263976450¬humb=yest_blank3.jpg(37.34KB)2009-5-1219:50classperf(cp,classes,test);cp.CorrectRateans=0.9867%分類器效果測評,就是看測試集分類的準確率的高低.二.臺灣林智仁的libsvm工具箱該工具箱下載libsvm-mat-2.86-1:HYPERLINK/attachment.php?aid=24867&k=c05691426227073bc1909cb597eceacd&t=1263

8、976450t_blanklibsvm-mat-2.86-1.rar(73.75KB)HYPERLINK/attachment.php?aid=24867&k=c05691426227073bc1909cb597eceacd&t=1263976450t_blanklibsvm-mat-2.86-1.rar(73.75KB)下載次數(shù):3732009-5-1220:02安裝方法也很簡單,解壓文件,把當前工作目錄調(diào)整到libsvm所在的文件夾下,再在setpath里將libsvm所在的文件夾加到里面.然后在命令行里輸入mex-setup%選擇一下編譯器make這樣就可以了.建議大家使用libsvm工

9、具箱,這個更好用一些.可以進行分類多類別,預測.=svmtrainsvmpredict=簡要語法:Usage=matlabmodel=svmtrain(training_label_vector,training_instance_matrix,libsvm_options);-training_label_vector:Anmby1vectoroftraininglabels(typemustbedouble).-training_instance_matrix:Anmbynmatrixofmtraininginstanceswithnfeatures.Itcanbedenseorspars

10、e(typemustbedouble).-libsvm_options:AstringoftrainingoptionsinthesameformatasthatofLIBSVM.matlabpredicted_label,accuracy,decision_values/prob_estimates=svmpredict(testing_label_vector,testing_instance_matrix,model,libsvm_options);-testing_label_vector:Anmby1vectorofpredictionlabels.Iflabelsoftestdat

11、aareunknown,simplyuseanyrandomvalues.(typemustbedouble)-testing_instance_matrix:Anmbynmatrixofmtestinginstanceswithnfeatures.Itcanbedenseorsparse.(typemustbedouble)-model:Theoutputofsvmtrain.-libsvm_options:AstringoftestingoptionsinthesameformatasthatofLIBSVM.ReturnedModelStructure=實例研究:loadheart_sc

12、ale.mat%工具箱里自帶的數(shù)據(jù)如圖:tu4HYPERLINK/attachment.php?aid=24873&k=f40a408143396519121ed2f478a12159&t=1263976450¬humb=yest_blank4.jpg(9.36KB)2009-5-1220:08其中heart_scale_inst是樣本,heart_scale_label是樣本標簽model=svmtrain(heart_scale_label,heart_scale_inst,-c1-g0.07);%訓練樣本,具體參數(shù)的調(diào)整請看幫助文件predict_label,accuracy,de

13、c_values=svmpredict(heart_scale_label,heart_scale_inst,model);%分類預測,這里把訓練集當作測試集,驗證效果如下:predict_label,accuracy,dec_values=svmpredict(heart_scale_label,heart_scale_inst,model);%testthetrainingdataAccuracy=86.6667%(234/270)(classification)=這回把SVM這點入門的東西都說完了,大家可以參照著上手了,有關SVM的原理我下面有個簡易的PPT,是以前做項目時我做的當時我負

14、責有關SVM這一塊代碼實現(xiàn)講解什么的,感興趣的你可以看看,都是上手較快的東西,想要深入學習SVM,你的學習統(tǒng)計學習理論什么的.挺多的呢.HYPERLINK/attachment.php?aid=24876&k=fe23ad40b82e1b2d910e3bd90fc2c6bb&t=1263976450t_blankSVM.ppt(391KB)HYPERLINK/attachment.php?aid=24876&k=fe23ad40b82e1b2d910e3bd90fc2c6bb&t=1263976450t_blankSVM.ppt(391KB)下載次數(shù):4292009-5-1220:18-有關S

15、VM和libsvm的非常好的資料,想要詳細研究SVM看這個-HYPERLINK/attachment.php?aid=32035&k=7c402be507a1bc9299a7f8969efef751&t=1263976450t_blanklibsvm_guide.pdf(194.53KB)HYPERLINK/attachment.php?aid=32035&k=7c402be507a1bc9299a7f8969efef751&t=1263976450t_blanklibsvm_guide.pdf(194.53KB)下載次數(shù):1862009-8-1914:58HYPERLINK/attachment.php?aid=32036&k=5a148648ce8445a95e0c63d9588a0dad&t=1263976450t_blanklibsvm_library.pdf(316.82KB)HYPERLINK/attachment.php?aid