[pymvpa] Spam No.3 (Bug?!)

[James M. Hughes]

Hi all,

As you may have guessed, there's something of a recurring theme here  
today.

I have a general question about the SVM implementation and then a  
concern.  My question:

Is it the case that we can use any SVM w/ multi-class data, and that  
the SVM implementation will handle this case correctly, or do we have  
to do pair-wise splits beforehand to reduce it to a two-class problem?

The reason I ask (and am concerned) is that I have written a project  
which uses PyMVPA.  Basically, I'm taking one of the Haxby Eight  
subjects, importing the data, then selection a random subset of 12000  
of the original voxels.  I want the students to perform classification  
initially using an SVM, then do feature selection and classify, to see  
the differences in accuracy that can be obtained using this method.   
Now, while the feature selection seems to work, students have reported  
that they get the same classification error using a Linear C-SVM, even  
when they drastically change the C value (from 10^-7 all the way to  
10^10).  Furthermore, the accuracy was the same for two different  
random data subsets.  This sounds bad to me, but I'm not exactly sure  
why, and I thought that the two vs. many classes issue might be a  
source of problems.

Any help would be greatly appreciated!

Thanks,
James.