tabsc - 2007-10-01

I'd like to use the AlphaY and SVs parameters to compute the decision function of each classifier in the case of a multiclass SVM (since it seems to me that OSU_SVM use the "one against all" strategie). Unfortunatly, the data is organized in a pretty fuzzy way (as said in code comments). Can someone help me with those coefficients ? Here are the comments :

%                All the AlphaYs are organized as follows: (pretty fuzzy !)
%                      classifier between class i and j: coefficients with
%                           i are in AlphaY(j-1, start_Pos_of_i:(start_Pos_of_i+1)-1),
%                         j are in AlphaY(i, start_Pos_of_j:(start_Pos_of_j+1)-1)
%    SVs       - Support Vectors. (Sample corresponding the non-zero Alpha), M x sum(nSV),

what I don't understand is that for a 3 class SVM I obtain only two rows... what is start_Pos_of_j or i (linked to nSV ?)
if j=1 how is it possible to have j-1 ?

I'd like those parameters to be able to compute posterior probabilities for SVM using Platt's method. This works really well for 2 classes so that it will be very interested to test it on my 3 classes problem.

I use OSU_SVM 3.0,

Thanks in advance,

tabsc