Hi,

I understand there are two methods implemented in shogun to determine
the subkernel weights during MKL classification training, and that
set_interleaved_optimization_enabled() is the interface to choose
between the two.

As far as I see it, interleaved optimization is supposed to be faster
but requires SVMLight to be used as a backend.

Among the undocumented examples in the repository, there is an example
named "classifier_mkl_svmlight_modelselection_bug.cpp" which contains

>         /* works */
>         CMKLClassification* classifier=new CMKLClassification(new CLibSVM());
>         classifier->set_interleaved_optimization_enabled(false);
> 
>         /* the above plus this does not work (interleaved only with 
> SVMLight)*/
> //      classifier->set_interleaved_optimization_enabled(true);
> 
>         /* However, SVMLight does not work */
> //      CMKLClassification* classifier=new CMKLClassification(new 
> CSVMLight());
> //       /* any of those */
> //      classifier->set_interleaved_optimization_enabled(false);
> //      classifier->set_interleaved_optimization_enabled(true);

Is this still valid? Is there currently any way to use the interleaved
method for mkl classification in shogun?

Cheers,
  Leon

-- 
Leon Kuchenbecker
Berlin-Brandenburg Center for Regenerative Therapies (BCRT)
Charité Universitätsmedizin
Föhrer Straße 15
13353 Berlin


Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to