How to combine binary classifier's results to output multi-class classification adopting File Exchange example in Libsvm's precomputed svmtrain?
4 views (last 30 days)
In File Exchange example: https://www.mathworks.com/matlabcentral/fileexchange/39352-multi-class-svm under %build model heading, author has the following line of code:
models(k) = svmtrain(TrainingSet,G1vAll);
How to use the same approach using above line of code with Libsvm's precomputed kernel? In particular, how to use it here:
model = svmtrain(TrainLabels, double(K), '-t 4');
Note that here svmtrain is a Libsvm function not to be confused with MATLAB's svmtrain function.
I have used Libsvm's precomputed kernel for binary classification using one-vs-one approach. Each one of these binary classification results give output accuracies. I will like to combine/ensemble all these accuracies to get one final output accuracy equivalent to that of multi-class classifier. For this reason, I looked for above File Exchange example which works fine for MATLAB's svmtrain function, but I am not sure how to use the same approach for LIBSVM's precomputed kernel function using their svmtrain function. Perhaps there is other way to obtain my desired result.