No one has replied to the message below,
Thank you very much for your reply, but I am already doing that for SMO and I have no problem,
But what you say does not work for SMOReg? Weka says: "can not handle numeric attributes. "
A simple arff and precomputed kernel file is below.
What do you suggest?
@attribute index numeric
@attribute target numeric
On 02/23/2012 09:54 AM, ahmet okutan wrote:
> Hi all,
> I am trying to use Weka with SMOreg and a precomputed kernel matrix that
> includes similarity values.
> But each time I receive an error where the weka classifier says: can not
> handle numeric attributes.
> I believe that it is possible to use SMOreg with numeric values, is not it?
> But it is not possible in my case, what shall I do to make SMOreg be
> able to use numeric attributes?
The input ARFF file that you use with SMO and the precomputed kernel
matrix needs to have two attributes - the first is an index for the
rows/columns of the matrix and the second is the class attribute. The
index attribute must be declared as nominal. We should probably allow
the index attribute to be numeric too, but this would require that the
user remembers to turn off normalization in SMO.