v0.4

First version with Run3 included. Main changes:

• adds configs c22pre and c22post
• adds new default Selectors for SL (sl1) and DL (dl1)
• adds categories for ABCD
• lots of configuration changes

v0.3

Last merge before switching efforts to include Run 3 into the analysis.

Main changes:

include resonant analysis (#59)
implement noise filters, json filter, primary vertex requirements (#60)
custom plotting functions: S over B (#61) and cutflow S over B (#63)
bjet regression Calibrator as well as plotting functions for fitting (#64)
GenParticle categorizaton (#65)
Neutrino + Top mass reconstruction (#66)

v0.2

• include DL channel up to datacards
• improvements to MLModel and InferenceModel
• add scripts to allow working with notebooks
• columnflow updates

For collaborators from UHH, we also provide a first reference result, which can be used by using the law.shared.cfg as the law config. It's good practice to write your own law config, which inherits from the law.shared.cfg. Such a law config could look like this:

[core]

inherit: $HBW_BASE/law.shared.cfg

[analysis]

# set your default analysis inst to "hbw_sl"
default_analysis: hbw.analysis.hbw_sl.hbw_sl
default_config: c17
default_dataset: ggHH_kl_1_kt_1_sl_hbbhww_powheg


[outputs]
# decide, for which tasks you want to produce your own outputs, e.g. rerun the MLTraining and everything that follows

cf.BundleRepo: local
cf.BundleSoftware: local
cf.BundleBashSandbox: local
cf.BundleCMSSWSandbox: local
cf.BundleExternalFiles: local

cf.ProduceColumns: local
cf.MLTraining: local
cf.MLEvaluation: local
cf.CreateHistograms: local
cf.MergeHistograms: local
cf.MergeShiftedHistograms: local
cf.PlotCutflow: local
cf.PlotCutflowVariables: local
cf.PlotVariables: local
cf.PlotShiftedVariables: local
cf.CreateDatacards: local

v0.1

This release provides a first refererence result for the HH->bbWW(SL) analysis. The full analysis chain from NanoAODs up to combine datacards is implemented and has been tested.
The main analysis steps include:

• JEC/JER calibrations
• object and event selection
• Machine Learning workflow (including input preparation, training, evaluating, and plotting)
• Production of event weights (pdf, scale, pu, btag, e/mu corrections)
• extensive categorisation based on leptons, jets, bjets, and the ML output nodes
• producing histograms, plots and datacards
• rebinning histograms used in the datacards

The main issue with this version of the repository is that the performance of the MLTraining seems to be rather poor, which results in rather bad sensitivity in the final combine fit.

The full workflow up to rebinned datacards can be called with the command:

law run hbw.ModifyDatacardsFlatRebin --version prod1 --inference-model rates_only

However, this will start the workflow with only a single local worker and will therefore take forever. To prevent this, you can rely on the hbw_datacards function, which calls the CreateDatacards with some parameters to run tasks via job submisson on htcondor (also using GPUs for the MLTraining task)

cd $HBW_BASE/scripts
source hbwtasks.sh
hbw_ml_training