I can not use libsvm!!!

Hi,

can anybody help me in using libsvm?

but it can not be recognized by my matlab, I got the following error:

??? Undefined function or method 'libsvmtrain' for input arguments of type 'double'.

Error in ==> svmtrts at 139 D.net.svm=libsvmtrain(otY,stX,'-t 0');

Error in ==> main at 148 [D,Dtest]=svmtrts(trndataSVM,tstdataSVM,'libsvm');

can anybody help me.

Regards, Atieh

Answer 1

[Friedrich]

I think your are talking about:

http://www.csie.ntu.edu.tw/~cjlin/libsvm/

The function libsvmtrain does not exist in that package. The training function is called svmtrain.

Answer 2

[Jonas Reber]

I used libsvm myself in matlab.

• download it from ( http://www.csie.ntu.edu.tw/~cjlin/libsvm/#download)
• add the svmtrain, svmpredict, libsvmwrite & libscmread .mex Files to your matlab path (probably you just put them in the working folder...)

then let me provide you my sample code - note: I use precompiled kernel data.

Here, I would like to find the optimal parameter c for my SVM.

    clear all; close all; 

 %%load datasets
    [lvtest,  test]  = libsvmread('test.krnl');
    [lvtrain, train] = libsvmread('train.krnl');
    [lvvalid, valid] = libsvmread('valid.krnl');

 %%optimize parameter c on validation set
    n = -17:17;
    accuracy = nan(size(n));

    for i=1:numel(n);   % n = {-17,...,17}
        c=2^n(i);       
        % create model
        model = svmtrain(lvtrain, train,['-q -t 4 -c ' num2str(c)]); 
            % option: -t 4 -> precomputed kernel
        [lbl, acc, dec] = svmpredict(lvvalid, valid, model);
        accuracy(i) = acc(1);
    end

    % output the accuracy vs the chosen parameter c
    plot(accuracy);
    xlabel('c'), ylabel('Accuracy'); title('Accuracy vs. c');

 %%test optimal c on the test set
    [~, i] = max(accuracy); % find the best value
    c = 2^n(i);             % this is the optimal c

    % create model
    model = svmtrain(lvtrain, train,['-q -t 4 -c ' num2str(c)]); 

    % test on the testset
    [lbl, acc, dec] = svmpredict(lvtest, test, model, []);

    % show accuracy
    disp(['Accuracy with optimized c (' ...
          num2str(c) ') on Testset: ' num2str(acc(1)) '%']);

hope this helps...?

[judy frost]

Can you please explain the example further by showing how to find optimal cost and gamma values from validation data that are used for k-fold cross validation. Furthermore is it possible to plot the graph of classified data at the end of validation,training and test stages. I will appreciate any further explanation regarding the topic. Thank you.

[Najiyah Valappil]

Can you plaese explain how this data was prepared?