TMVA set weights for S and B

TMVA
1

Dear Experts,

I’m using TMVA in order to optimize a selection.
I set the weights for Signal and Backgrounds with:
factory.AddSignalTree ( signal, signalWeight )
factory.AddBackgroundTree( background, backgroundWeight )
Where signalWeight and backgroundWeight are the weight for the S and B, given by the cross section and the number of events generated in the MC.
Then, when I look at the output from TMVAGui.C I see that:
a) All the plots like mva_LikelihoodD.png are normalized to same area (but I know this is normal).
b) The plots like mvaeffs_BDT.png have written a value for S/sqrt(S/B) that is unrealistic, since here the relative normalization has not been taken into account.

The efficiencies for S and B are independent from the relative normalization, but the purity, the signal efficiency*purity and the s/sqrt(S+B) values make no sense without the relative normalization.

So I ask:

  1. How can I be sure that the normalization has been taken into account?
  2. How can I have plots where the normalization is taken into account?

Thanks a lot,
Luca

1 Like
2

Hi Luca,

It would be good to redirect your question to the TMVA forum, where you can get answers from both TMVA experts and users:

tmva-users@lists.sourceforge.net
sourceforge.net/p/tmva/mailman/tmva-users/

In particular, you may want to have a look at Peter’s answer to this question:
sourceforge.net/p/tmva/mailman/message/34431489/

Cheers,

Sergei