Re: [TMVA-users] a connection between xml file and rootfile

[Helge V. <Hel...@ce...>]

Hi Yuto,


On 22 September 2015 at 04:26, Yuto MINAMI <yut...@ce...> wrote:
> Dear experts,
>
>   I'm not sure this is the correct place to ask about TMVA usage.
> I'm sorry If there are other places to ask questions.
>
> I classified cuts of variables with modified TMVAClassification.cxx ,
> with the method "CutsGA".
> And I tried to get the cuts which make the significance max.
>
> I got the signal efficiency from "MVA_CutsGA_effS" histogram and BG
> efficiency from "MVA_CutsGA_effB".

Note, normally (for all classifiers BUT Cuts), where you have only one finall
output variable (the MVA value) the x-axis of these plots are the MVA-cut value
and the y-axis is the correspoingin efficiency. For Cuts, this isn't
possible, hence
we replaced the x-axis by the 'signal effiency' (which you have to
translate yourself
to the corresponding set of cuts.

As a result,  in the MVA_CutsGA_effS plot, you see a 'nice' diagonal,
as x-axis and y-axis
are equivalent (apart from some 'numericsal differences')

Now I don't see you you can 'read off' any efficiency from that plot ?

The only think you can use is" MVA_CutsGA_effB" which will tell you
for which signal
effieicney (x-axis) you would get which kind of background efficiency (y-axis)


> Then I calculated significance at each bin and get the maxmum
> significant signal efficiency.
> With the signal efficiency, I got the cut value from the xml file with
> "TMVA::MethodCuts::GetCuts" function.
>
> Then I have two questions.
>
> 1. Which signal efficiency should I use to get cuts ?
>      X axis or Yaxis value  of MVA_CutsGS_effS ?

See above, they are apart from bining/numerical differences equivalent ..

>
> 2. Is bin content of "MVA_CutsGA_effB" correspond to the "effB" in xml
> file ?
>     I checked the each bin of "MVA_CutsGS_effB" and "effB", and in many
> case they show similar value.
>      However, sometimes , they show very different value.
>      Is this the correct behavior?

hmm.. they should be the same, differences I could imagine come only
about through
possible smoothing of histogram etc... I 'd have to check in the code.
But what do you mean by
"Very different" value ??

Cheers,

Helge


>
>
> I used TMVA-v4.2.0 for this study.
>
>
> Sincerely,
> Yuto
>
> --
> orz orz orz orz orz orz orz orz orz orz orz orz
> Yuto MINAMI (yut...@ce...)
> TEL  : +41-22-76-75288 (75288 at CERN)
> FAX  : +41-22-767-9750
> ROOM : Build 188/5-015
> A27100, CERN CH-1211, Geneva 23, Switzerland
>
>
> ------------------------------------------------------------------------------
> _______________________________________________
> TMVA-users mailing list
> TMV...@li...
> https://lists.sourceforge.net/lists/listinfo/tmva-users