On the \xi\alpha and GACV for tuning SVM's

[Grace Wahba]

 Paper available: "On the Relation Between the GACV and 
 Joachims' \xi\alpha Method for Tuning Support Vector
 Machines, With Extensions to the Nonstandard Case". 
                 by
 Grace Wahba, Yi Lin, Yoonkyung Lee and Hao Zhang
          UW-Madison Statistics TR 1039

 ftp://ftp.stat.wisc.edu/pub/wahba/xan.ps

              Abstract

We rederive a form of Joachims' $\xi\alpha$ method for 
tuning Support Vector Machines by the same approach as 
was used to derive the GACV, and show how the two
methods are related. We generalize the $\si\alpha$ 
method to the nonstandard case of nonrepresentative training
set and unequal misclassification costs and compare
the results to the GACV estimate for the standard and nonstandard
cases.