Hi all,
I want to use BDT to distinguish signal and background . For example , the proportion of signal: background is 1:10^4 . If I pass , for example 100 signal and 10^6 background and ask BDT to distinguish them, I worry that in this situation the result is invalid .
So I guess I should use Weight , for example , input both 10^6 signal and background , and add a weight = 10^-4 for signal , Am I right ?
Maybe I have asked once , I want to make the question clearly.
Best,
Li
I am a little confused . Is my question full of confusion or what? I think there are many experts, and many algorithm include the conception of weight. I can not understand why ’ weight ’ can improve us a right result? I think it is interesting. No body is interested in it?
Or, if you do not use weight, how can you deal with this case: for example , you got 100 signal and 10^6 background?
Hi Li,
As you want to best separate signal from background, I suggest that you start with an equal proportion of signal and background events during training. To get a better feel for what happens as you vary this ratio, try it for a few values and evaluate the resulting classifiers in the expected regime of signal and background.
Best regards,
Sergei
[quote=“sergei”]Hi Li,
As you want to best separate signal from background, I suggest that you start with an equal proportion of signal and background events during training. To get a better feel for what happens as you vary this ratio, try it for a few values and evaluate the resulting classifiers in the expected regime of signal and background.
Best regards,
Sergei[/quote]
Hi Sergei,
Thanks very much, I think your suggestion is practical. I think the change of ratio can tell me which parameter is important and give me some clues.
Best,
Li