Is there an easy way to optimize variables in TMVA?

TMVA
1

Hi all,

I want to optimize my input variables for BDT as removing them one by one and check their performance by ROC curve for each step. I know this method is kind of a common sense in optimizing the variables, so I wonder if there is somehow an easy way to do it, i.e. a script already implemented in TMVA/ROOT interface etc.?

Thanks a lot.

2

Hi,

Not that I am aware of. If you have many variables maybe you can gain insight (and subsequently some speed) by checking the variable importance output after the BDT training.

Cheers,
Kim

3

“Gene Importance” might be something you’re looking for.
Also covariance matrix may help.

But in the end, I don’t think there are well definition of what are the best variables.