Just having been made aware of the useful optimisation nodes in Thorsten's UGM talk, I have an immediate feature request: could we brute-force exponentially as the classic SVM primer paper suggests? It's probably a preferable brute-force stategy for most use cases beyond SVM as well.
Hmm, do you have a citation for the paper? A quick google search didn't yield anything and I am not already familiar with the method.
Thanks again for coming to see us in Zurich!
Sorry, missed your reply. It's "A Practical Guide to Support Vector Classification" from the libSVM site over here:
Section 3.2 / page 5:
"We recommend a grid-search on C and ɣ using cross-validation. Various pairsof (C, ɣ) values are tried and the one with the best cross-validation accuracy is picked. We found that trying exponentially growing sequences of C and ɣ is a practical method to identify good parameters"