I need to get the posterior probabilities output of trained SVM instead of the binarized output. In the latest versions of Matlab, this can be done by the following steps:
SVMModel = fitcsvm(trainingInput,trainingTarget);
SVMModelPosteriorProb = fitSVMPosterior(SVMModel);
[label, probability] = predict(SVMModelPosteriorProb,testInput);
However, fitcsvm and fitSVMPosterior are both introduced in R2014a version and I am using an earlier one. When I pass the output of the svmtrain function to the predict function, I get an error message stating that struct is not the expected argument type.
Is there a way to do this in earlier Matlab versions?
There are lots of MATLAB codes and toolboxes available online for either svm or rvm. using them you will not need MTALAB higher version. if you need posterior probability why don't you use RVM instead of SVM?
Regards;
Shiva
Hi Shiva,
when I was working with SVM I used the Libsvm library, for which you can also find compiled executables that can be simply called by matlab (https://www.csie.ntu.edu.tw/~cjlin/libsvm/).
Libsvm also provides the possibility to compute the Platt's probabilistic output. Otherwise you can try to implement yourself this algorithm: http://www.csie.ntu.edu.tw/~htlin/paper/doc/plattprob.pdf
Good luck,
Loris
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining