TMVA variables include/exclude

[Geo]

Hello

I was wondering if I can exclude some variables during the training and testing phase, but include them in the application phase.
Does something like that makes even sense?

I don’t mean as spectators.

cheers

[kialbert]

Hi George!

This is not possible to do in TMVA and not possible for MVA methods in general. The variables used in a training is often deeply tied to how that classifier does its classification, so one cannot be excluded and still produce sensible results.

What is it you would want to achieve by doing this?

Cheers,
Kim

[Geo]

Hello and thank you for the answer. I actually was wondering if something like that makes even sense. I made this topic because I couldn’t find anywhere any specific answer about that.

What I wanted was to keep out of training-testing an important for my analysis variable in order to see afterwards to what extent the tmva can do a proper job.

thank you very much

George

[kialbert]

What you could do is two separate trainings, one with the variable and one without and compare the classification performance of those two cases.

You can also look at the variable ranking output of TMVA (for supported method, output when you run EvaluateAllMethods) to get an idea of how much a variable affects the final classification.

Cheers,
Kim