Tag Archives: Computing

Multivariate Salad Bar

For those of you interested in pattern classification, machine learning, etc, there’s a really nice package called TMVA, which contains a ton of multivariate techniques which are fairly easy to integrate into High Energy Analyses, as they are a ROOT extension. Even if you don’t do HEP, they’re nice because all the methods have comparable inputs, so you can test tons of different ones without a big development hit. I’ve only futzed with the Multi-Layer Perception package, but that was pretty quick and painless to implement. A buddy of mine at H1 used it for electron tagging, and tested like seven different methods before settling on a Boosted Decision Tree. I mention this now, because Anselm Vossen just released a couple of lecture notes (I think they’re from the CERN Computing School) on the principles of some of the implemented techniques specifically Support Vector Machines and one which has a nice discussion of data preprocessing for MVA. If your shopping for selection criteria, check em out.

That reminds me, a reader “A. Game Coder” commented before that they’d like to see more posts on computing in Physics. While I’m thinking about it, check out the slides from the CERN summer school in 2006 and 2005. Lots of fun stuff like genetic algortihms and management of large data sets.