TMVA: Uncertainty for regression value in BDT

TMVA
1

I am using boosted decision trees (as a regressor) in TMVA. I want an error estimate along with the prediction (e.g. EvaluateMVA() --> 2.5 ± 0.5). I’ve combed through the documentation and I thought that perhaps Reader->GetMVAError() was what I needed but this always returns exactly zero (after calling EvaluateMVA first).

Is there a flag that needs to be set during training to get access to this feature? Or is this not the intent of this method?

Thank you.

2

Hi,

sorry for the delay, I’m not following the ‘roottalk’ but rather the TMVA mailing list for TMVA related questions. Anyway, you are right, the Reader->GetMVAError() would be the natural thing to look at, however, error estimates are not implemented for all ‘regressors’ (in fact only for very few) and BDT is not amongst them.

Cheers,

Helge