One question about variable ranking in TMVA package

TMVA
1

Hi all
I recently notice one tiny difference about input variables ranking between method unspecific and method specific.
if I do not assign the classfication method .The sum of all input variables’ weights are not normalizing to 1.As is shown in figure01.
Conversely，if I assign one method such as BDT .The sum of all input variables’ weights are normalizing to 1.As is shown in figure02.
I just wonder why we do not make a normalization in method unspecific situation.Besides,I would like to know the reason why the input variables ranking is different among different classfication method.Thanks so much!!

rank01
rank01760×300 24.6 KB

rank02
rank02703×304 24.1 KB

2

Hi @melodynk ,
I think @moneta might be able to help, let’s ping him.

Cheers,
Enrico

3

HI,
This input variable ranking is based on different method so it is different.
The first one is based on the separation index of the input variables, see equation (1), page 25 of

While the second one results from the BDT , see 8.13.4 in teh above Users Guide.

Cheers

Lorenzo