co-training.zip_co-training_co-training matlab_cotraining

clc; clear; %% 导入数据 data=importdata('mfeat-kar'); for i=0:9 for j=1:200 data(i*200+j:((i+1)*200),65)=i; end end %% 随机划分训练集和测试集 N=size(data); ind = randperm(N(1)); %产生一个随机的排序 sample_train=data(ind(1:N(1)*0.5),1:65); %随机挑选出训练样本, 50% sample_test=data(ind(N(1)*0.5+1:end),1:65); %随机挑选出测试样本, 50% %% 将训练集划分为有标记的数据集L和无标记的数据集U L/U=90% sample_train_L=sample_train(1:100,:); %得到有标记训练样本占训练集10% sample_train_U=sample_train(101:end,1:64); %得到无标记训练样本占训练集90% L1=zeros(1000,33); L2=zeros(1000,33); %将L分为两个视图L1 和 L2 L1(1:100,1:32)=sample_train_L(1:100,1:32); L1(1:100,33)=sample_train_L(1:100,65); L2(1:100,1:32)=sample_train_L(1:100,33:64); L2(1:100,33)=sample_train_L(1:100,65); for i=1:9 U11=sample_train_U(((i-1)*100+1):(i*100),1:32); % 得到一个初始的无标记子集U' meas1=L1(1:(i*100),1:32); species1=L1(1:(i*100),33); by1=fitcnb(meas1,species1); R1=by1.predict(U11); result1(1:100,1:32)=L1(1:100,1:32); result1(1:100,33)=R1(1:100,1); U12=sample_train_U(((i-1)*100+1):(i*100),33:64); meas2=L2(1:(i*100),1:32); species2=L2(1:(i*100),33); by2=fitcnb(meas2,species2); R2=by2.predict(U12); result2(1:100,1:32)=L2(1:100,1:32); result2(1:100,33)=R2(1:100,1); L1((i*100+1):((i+1)*100),:)=result2(1:100,:); L2((i*100+1):((i+1)*100),:)=result1(1:100,:); end %% h1对测试集进行分类 sample_test_t1=sample_test(:,1:32); %测试集的样本特征 sample_test_t2=sample_test(:,33:64); sample_test_c=sample_test(:,65); %测试集的类别 last_r1=by1.predict(sample_test_t1); cMat1=confusionmat(sample_test_c,last_r1); r=diag(cMat1); accuracy1=(sum(r))/1000; %% h2对测试集进行分类 last_r2=by2.predict(sample_test_t2); cMat2=confusionmat(sample_test_c,last_r2); r2=diag(cMat2); accuracy2=(sum(r2))/2000; disp('accuracy= '); if(accuracy1>=accuracy2) disp(accuracy1); else disp(accuracy2); end