Options
"Getting the function value of SVM as output [SOLVED]"
Contributor II
Good evening,
the more I am getting into RM, the more I like it. It is really a great piece of software.
Yet, the more I learn, the more questions I have. This time it is about support vector machines (SVM) for a binominal classification.
A SVM calculates weights and identifies support vectors in the learning stage. To classify a vector, the SVM applies the vector to a function that basically contains the Kernel function, the support vectors, and the weights. Depending on the sign of the resulting function value, the vector is classified as 0 or 1 (or whatever name you use for your classes).
Applying a SVM (LibSVM) in RM, I get the prediction and the confidence as output.
Is there also an easy (=integrated in the GUI) way to get the function value as output or do I have to modify the source code of the particular operator?
I am looking forward to your answer.
Best regards,
Alex
the more I am getting into RM, the more I like it. It is really a great piece of software.
Yet, the more I learn, the more questions I have. This time it is about support vector machines (SVM) for a binominal classification.
A SVM calculates weights and identifies support vectors in the learning stage. To classify a vector, the SVM applies the vector to a function that basically contains the Kernel function, the support vectors, and the weights. Depending on the sign of the resulting function value, the vector is classified as 0 or 1 (or whatever name you use for your classes).
Applying a SVM (LibSVM) in RM, I get the prediction and the confidence as output.
Is there also an easy (=integrated in the GUI) way to get the function value as output or do I have to modify the source code of the particular operator?
I am looking forward to your answer.
Best regards,
Alex
Tagged:
0
Answers
as far as I know this isn't possible at the moment. :-[
Best,
Nils
thanks for your reply. I've just manipulated the performPrediction method in com.rapidminer.operator.learner.functions.kernel.LibSVMModel.java. Instead of the confidence attribute, I am getting the function value now. This works for me.
Greetings,
Alex