Hello,

I have been able to convert the cars.data.arff file to the format needed by the libsvm library.  This is where I am stuck however.  I have read through the source code, the README file and the python files but I am still not sure how to proceed.

From the REAME file it seems like you first have to create a svm_problem object and an svm_parameter object in order to create the
svm_model object needed for further processing. Is this correct?

Now when creating the svm_parameter object:

We have to test each of the 5 svm types: C_SVC, NU_SVC, ONE_CLASS, EPSILON_SVR, NU_SVR?

And for each of those we need to try the 4 kernel types: LINEAR, POLY, RBF, SIGMOID?

Once the svm type and kernel type is chosen we have to test those choices with the various options for the variables within each kernel?  How do you decide what the best choice for the variables in each kernel are, as there seem to be an infinite number of combinations?  The beginners guide suggests for RBF for example to use a grid-search on the possible variable values using cross-validation and various combinations of the possible values are paired up and tested and the pair with the best cross-validation accuracy is the one that is picked. 

It seems like this would take a a lot longer than the 10 minute time limit if you use all 5 svm types, and for those all 4 kernels, and for those varying numerical options for the kernel variables.

As you can see I am quite confused as to how I should proceed.  As mentioned before I have read through all of the relevant source and example code files and the online "beginner's guide" as well as the included README file, and I am still quite confused.  Is anyone else in the same position?