svm.m

核方法＆svm是模式识别是很重要的方法

function [recRate,score,TP,TN,FP,FN] = SVM (trainData,trainLable,testData,testLable,trainNum,testNum,...
    kPar,C,kType)
%SVM: support vector machine
%input:
%  trainData: data for training
%  trainLable: training data's lable
%  testData: data for testing
%  testLable: testing data's lable
%  trainNum: the former trainNum trainData are malignant,the others are
%            benign
%  testNum: the former testNum testData are malignant,the others are benign
%  kPar: kernel function's parameter
%  C: punish parameter
%  kType: kernel function's type
%output:
%  recRate: the recognize rate
%  score: to produce ROC curve
%  TP: true malignant
%  TN: true benign
%  FP: false malignant
%  FN: false benign

length_train = size(trainData,2);
length_test = size(testData,2);
TP = 0;TN = 0;FP = 0;FN = 0;

trainData = trainData';
testData = testData';
Y = trainLable';
trainMalignant = trainData(1:trainNum,:);
trainBenign = trainData(trainNum+1:length_train,:);
% testMalignant = testData(1:testNum,:);
% testBenign = testData(testNum+1:length_test,:);

parameter = SVC_Fun(trainMalignant,trainBenign,trainNum,length_train-trainNum,kPar,C,kType);
index = find(parameter.alpha>0);
sup_vector = trainData(index,:);
mat_kernel = Kernel(testData,sup_vector,kType,kPar);
f_SVM = (mat_kernel*(parameter.alpha(index,:).*Y(index,:)))+parameter.b;

YY = sign(f_SVM);
for i=1:length_test
    if YY(i)==0
        YY(i) = 1;
    end
end
clear i;
for i=1:length_test
    if YY(i) == testLable(i)
        if i<=testNum
            TP = TP+1;
        else
            TN = TN+1;
        end
    else 
        if i<=testNum
            FN = FN+1;
        else
            FP = FP+1;
        end
    end
end
score = f_SVM';
recRate = (TP+TN)/length_test;