Home | News | Documentation | Download

Training TMVA with different signal/background variable names

Is it possible to train a classifier using two tuples (signal/background, MC/data) that have different variable names? For example, some variables are called PIDx in background, and PIDx_corr in the signal tree.

I guess I’m imagining something like TMVA::DataLoader::AddVariable with two TString expression arguments. But any workaround (without renaming the variables in the (large) tuples) would do. Thanks!

Hi @jruizvid; I am sure @moneta has the answer to your question.


Anybody knows if this is possible?
(reminder after few days with no answer)

No this is not possible, variable must have the same name, because what then will be called when using the trained method in the real un-labeled data ?
There is an easy solution, create a snapshot of the original tree with the variables renamed.