datasets with asymmetric errors - multiplicative uncertainties don't match additive

Created by: wilsonmr

Commondata files are layed out as per https://github.com/NNPDF/nnpdf/blob/master/doc/data/data_layout.pdf

It seems to me that we should have

100*(columns_additive/central_values) == columns_multiplicative

using https://github.com/NNPDF/nnpdf/pull/476 we can load data as a pandas dataframe and check this very easily and I find the following datasets to have some kind of conflict:

DATA_ATLASTOPDIFF8TEVTPTNORM.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_8TEV_TBAR_RAP_NORM.dat
DATA_ATLASTOPDIFF8TEVTTMNORM_SF.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_8TEV_TBAR_PT_NORM.dat
DATA_CMSJETS11_SF.dat
DATA_CDFR2KT_SF.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_7TEV_TBAR_RAP_NORM.dat
DATA_CMSJETS11.dat
DATA_ATLAS1JET11.dat
DATA_LHCBZEE2FB.dat
DATA_ATLASTOPDIFF8TEVTTMNORM.dat
DATA_ATLASPHT15_SF.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_8TEV_T_RAP_NORM.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_8TEV_T_RAP.dat
DATA_CMSTTBARTOT.dat
DATA_D0R2CON.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_7TEV_T_PT_NORM.dat
DATA_ATLASTOPDIFF8TEVTRAPNORM.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_8TEV_TBAR_RAP.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_8TEV_TBAR_PT.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_8TEV_T_PT_NORM.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_7TEV_TBAR_PT.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_7TEV_TBAR_PT_NORM.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_7TEV_TBAR_RAP.dat
DATA_D0ZRAP_SF.dat
DATA_ATLASPHT15.dat
DATA_D0ZRAP.dat
DATA_LHCBW36PB.dat
DATA_ATLASTOPDIFF8TEVTTRAPNORM_SF.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_7TEV_T_PT.dat
DATA_ATLASPHT12.dat
DATA_CMSTTBARTOT_SF.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_7TEV_T_RAP_NORM.dat
DATA_CDFR2KT.dat
DATA_ATLASTOPDIFF8TEVTRAPNORM_SF.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_7TEV_T_RAP.dat
DATA_ATLAS1JET11_SF.dat
DATA_ATLASTOPDIFF8TEVTTRAPNORM.dat
DATA_CDFZRAP.dat
DATA_ATLAS_SINGLETOP_TCH_DIFF_8TEV_T_PT.dat
DATA_ATLASTOPDIFF8TEVTPTNORM_SF.dat
DATA_ATLASPHT12_SF.dat

For some of these it might be a book keeping task because the uncertainties are additive. However take for example D0ZRAP which performs symmatrisation of errors, we see in

https://github.com/NNPDF/buildmaster/blob/12caaabafd824b6adc013eb205e121e25b0969cc/filters/D0.cc#L77

that the mult values are calculated with respect to the unshifted central values which I believe is incorrect. Some of the above datasets have this issue (and can be easily identified by the _SF suffix). For at least one other I have found a bug in the buildmaster implementation:

https://github.com/NNPDF/buildmaster/blob/12caaabafd824b6adc013eb205e121e25b0969cc/filters/LHCBZEE2FB.cc#L125

should read:

fSys[i][l].add  = syscor[i][l-1];
fSys[i][l].mult =  syscor[i][l-1].add/fData[i]*1e2;

(it is calculating the multiplicative unc. for the wrong systematic - this can be seen by examining the commondata)

I'm happy to go through more of these sets but perhaps I could have some assistance?