i 'm trying to use the minst data with the support vector machine (SVM) i have a problem with the time of execution because minist data have a large dimension 70k,how can i use the PCA to reduce these dimensions . test_set=[60 000*748];train_set=[10 000*748];train_label=[60 000*1] and test_label=[10 000*1] . thank you
clear all; load('train_set.mat') load('test_set.mat') %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%ù train_label=[0*ones(1,6000) 1*ones(1,6000) 2*ones(1,6000) 3*ones(1,6000) 4*ones(1,6000) 5*ones(1,6000) 6*ones(1,6000) 7*ones(1,6000) 8*ones(1,6000) 9*ones(1,6000)]; train_label1=train_label'; model = svmtrain(train_label1, train_set, '-s 0 -t 0'); %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% test_set=[0*ones(1,1000) 1*ones(1,1000) 2*ones(1,1000) 3*ones(1,1000) 4*ones(1,1000) 5*ones(1,1000) 6*ones(1,1000) 7*ones(1,1000) 8*ones(1,1000) 9*ones(1,1000) ]; test_set1=test_set'; [predicted_label, accuracy, decision_values]=svmpredict(test_label1, test_set, model); %%%%%%%%%%%% nt=1000; m=6000; [Pa,V,me1,l1] = pcaproj(train_set1,2); [n,p]=size(test_set1); Pt=(test_set1-ones(n,1)*me1)*V; min(Pa); max(Pa); min(Pt); max(Pt); m1 = -1; M1 = 2.5; m2 = -1.5; M2 = 1; ind1 = 0.25; ind2 = 0.25; [xg yg] = meshgrid([m1:ind1:M1],[m2:ind2:M2]); [nx,ny] = size(xg); X = [reshape(xg ,nx*ny,1) reshape(yg ,nx*ny,1)]; [n,p] = size(X); Xg = X*V' + ones(n,1)*me1; LabelS1=[1*ones(1,1000) 2*ones(1,1000) 3*ones(1,1000) 4*ones(1,1000) 5*ones(1,1000) 6*ones(1,1000) 7*ones(1,1000) 8*ones(1,1000) 9*ones(1,1000) 10*ones(1,1000) ]; [ypred, accuracy1, decision_values1]=svmpredict(LabelS1, Xg, model); decision_values1=decision_values1'; maxi=max(decision_values1); ypred = reshape(ypred,nx,ny); contourf(xg,yg,ypred);shading flat;hold on; [cs,h]=contour(xg,yg,ypred,[1 2 3]); clabel(cs,h); style= ['^+*']; color= ['kkk']; hold on; nbclass=max(train_label); for i=0:nbclass-1 h=plot(Pt(i*nt+1:(i+1)*nt,1),Pt(i*nt+1:(i+1)*nt,2),[style(i+1) color(i+1)]);hold on; end;
Dear researcher, try pcares method to reduce dimensions, it may work.
You can also go through the below link
http://in.mathworks.com/help/stats/princomp.html
In the newest version of MATLAB you can find the PCA method in classification learner app and and simply apply it when you want to train your model.
thank you @ Paolo Inglese , how can i apply this to my exemple ;
load('train_set.mat')
train_label=[0*ones(1,6000) 1*ones(1,6000) 2*ones(1,6000) 3*ones(1,6000) 4*ones(1,6000) 5*ones(1,6000) 6*ones(1,6000) 7*ones(1,6000) 8*ones(1,6000) 9*ones(1,6000)];
train_label1=train_label';
model = svmtrain(train_label1, train_set, '-s 0 -t 0');
- Badges
- Science topic
- Similar topics
- Mathematics
- Algebra
- Matrix