Validation Plots

Some validation plots for the CARL way of determining variation weights associated with Shower/Hadronization systematic uncertainties.

Link to the HZZ presentation: https://indico.cern.ch/event/1050099/

The NN will output a score s(x) for each event x, which can then be used to form the ratio variation as such:

var_ratio = (sigma_var/sigma_nominal) * (s(x)/(1-s(x))

The value of sigma_var/sigma_nominal for each of the variations is recorded below.

qqZZ samples

PATH to NN models giving the following diagnostics: /afs/cern.ch/user/j/jsandesa/public/forSam/arch_qqZZ (it contains the architecture, weight and variables.json file)

Note: The variables.json file in the path above contains the information on which variables were used for training as well as the scale/offsets for data preprocessing. Here is the nomenclature for some of the observables that required non-linear transformations to be used as inputs:

log_higgs_pt_fidBorn_truth_p2 : log(higgs_pt_fidBorn_truth+2.0)

log_higgs_m_fidBorn_truth : log(higgs_m_fidBorn_truth)

CKKW15 varaiton

sigma_CKKW15/sigma_nominal = 1.02 (inclusive)

sigma_CKKW15/sigma_nominal = 1.04 (selection m4l>180)

CKKW30 variation

sigma_CKKW30/sigma_nominal = 0.97 (inclusive)

sigma_CKKW30/sigma_nominal = 1.00 (selection m4l>180)

QSF4 variation

sigma_QSF4/sigma_nominal = 0.99 (inclusive)

sigma_QSF4/sigma_nominal = 1.01 (selection m4l>180)

QSF025 variation

sigma_QSF025/sigma_nominal = 1.01 (inclusive)

sigma_QSF025/sigma_nominal = 1.05 (selection m4l>180)

CSSKIN variation

No variation sample available

ggZZ SBI samples

PATH to NN models giving the following diagnostics: /afs/cern.ch/user/j/jsandesa/public/forSam/arch_ggZZSBI (it contains the architecture, weight and variables.json file) (In progress)

Note: The variables.json file in the path above contains the information on which variables were used for training as well as the scale/offsets for data preprocessing. Here is the nomenclature for some of the observables that required non-linear transformations to be used as inputs:

log_higgs_pt_fidBorn_truth_p2 : log(higgs_pt_fidBorn_truth+2.0)

log_higgs_m_fidBorn_truth : log(higgs_m_fidBorn_truth)

CKKW15 varaiton

sigma_CKKW15/sigma_nominal = 0.99 (inclusive)

sigma_CKKW15/sigma_nominal = 0.99 (selection m4l>180)

CKKW30 variation

sigma_CKKW30/sigma_nominal = 1.08 (inclusive)

sigma_CKKW30/sigma_nominal = 1.08 (selection m4l>180)

QSF4 variation

sigma_QSF4/sigma_nominal = 0.81 (inclusive)

sigma_QSF4/sigma_nominal = 0.81 (selection m4l>180)

QSF025 variation

sigma_QSF025/sigma_nominal = 1.38 (inclusive)

sigma_QSF025/sigma_nominal = 1.37 (selection m4l>180)

CSSKIN variation

sigma_CSSKIN/sigma_nominal = 1.02 (inclusive)

sigma_CSSKIN/sigma_nominal = 1.02 (selection m4l>180)

ggZZ Sig samples

PATH to NN models giving the following diagnostics: /afs/cern.ch/user/j/jsandesa/public/forSam/arch_ggZZSig (it contains the architecture, weight and variables.json file) (In progress)

Note: The variables.json file in the path above contains the information on which variables were used for training as well as the scale/offsets for data preprocessing. Here is the nomenclature for some of the observables that required non-linear transformations to be used as inputs:

log_higgs_pt_fidBorn_truth_p2 : log(higgs_pt_fidBorn_truth+2.0)

log_higgs_m_fidBorn_truth : log(higgs_m_fidBorn_truth)

CKKW15 varaiton

sigma_CKKW15/sigma_nominal = 1.012 (inclusive)

sigma_CKKW15/sigma_nominal = 1.004(selection m4l>180)

CKKW30 variation

sigma_CKKW30/sigma_nominal = 1.019 (inclusive)

sigma_CKKW30/sigma_nominal = 1.01 (selection m4l>180)

QSF025 variation

sigma_QSF4/sigma_nominal = 1.289 (inclusive)

sigma_QSF4/sigma_nominal = 1.278 (selection m4l>180)

CSSKIN variation

sigma_QSF4/sigma_nominal = 1.022 (inclusive)

sigma_QSF4/sigma_nominal = 1.013 (selection m4l>180)

VBF Samples

-- JayAjitbhaiSandesara - 2021-12-06