BDT training using signal weighted data (splot)

[RISHABH_RATURI1]

Hello everyone,

I’m trying to train the BDT classifier where the input background is the signal-weighted data using splot technique. After applying the signal-weight to the ntuple some the variables that are used as inputs to BDT have negative content and the TMVAClassification.C is giving error.

: error in matrix diagonalization; printed S and B
: error in matrix diagonalization; printed S and B
: Variable “Mumumass” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “Mumumass” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “Mumumass” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “Bmass” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “Bmass” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “Bmass” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “TMath::Max(MumMinIP,MupMinIP)” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “TMath::Max(MumMinIP,MupMinIP)” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “TMath::Max(MumMinIP,MupMinIP)” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “TMath::Max(Mumdcasigbs,Mupdcasigbs)” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “TMath::Max(Mumdcasigbs,Mupdcasigbs)” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “TMath::Max(Mumdcasigbs,Mupdcasigbs)” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “TMath::Max(Mumpt,Muppt)” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “TMath::Max(Mumpt,Muppt)” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
: Variable “TMath::Max(Mumpt,Muppt)” has zero, negative, or NaN RMS^2: -nan ==> set to zero. Please check the variable content
TFHandler_Factory : Variable Mean RMS [ Min Max ]
: -----------------------------------------------------------
: Mumumass: -nan 0.0000 [1.7977e+308-1.7977e+308 ]
: Bmass: -nan 0.0000 [1.7977e+308-1.7977e+308 ]
: IP: -nan 0.0000 [1.7977e+308-1.7977e+308 ]
: DCA: -nan 0.0000 [1.7977e+308-1.7977e+308 ]
: Pt: -nan 0.0000 [1.7977e+308-1.7977e+308 ]
: -----------------------------------------------------------
: -nan -nan -nan -nan -nan -nan
: signal and background histograms have different or invalid dimensions:
***> abort program execution
terminate called after throwing an instance of ‘std::runtime_error’
what(): FATAL error

Please suggest how can I train the BDT using this signal-weighted data.

[bellenot]

@moneta will most probably help you once back from vacation

[moneta]

Hi,

I think BDT supports negative weights, although I have no experience with those. The problem here is that the variance is negative. This should not be problematic if you are not using the rms in variable transformations. Are you using transformations ?
Probably yes, because I see that the resulting mean is a Nan.

Cheers

Lorenzo