VH

Setup

• source setup.sh
• law index

Tasks

Previous Tasks (usally already executed)

• DownloadFiles
• DownloadNanoAODs
• Corrections

Tasks to execute

The task chain runs as follows:

• CoffeaProcessor Processor
• GroupCoffeaProcessor
• Rebinning
• Plotting/Datacard

The DNN-training and exporting of variables to arrays is a different step:

• CoffeaProcessor - ExporterProcessor
• DNN-training tasks
• insert model-path

Law-tips Run all the following commands inside the common directory. When you execute a law task by law run TASKNAME --param1 --param2 ... you can add one of the following to check some properties of the task-chain. If you let for example the PlotProducer run and the CoffeaProcessor has not run yet, it will run automatically. Thus, you do not need to run all the commands seperately.

The --print-status X flag with X being an integer (0, 1, 2, ... or -1), shows you the status of tasks.

The --print-output X flag with X being an integer (0, 1, 2, ... or -1), shows you the output of tasks.

The --remove-output X flag with X being an integer (0, 1, 2, ... or -1), DELETES the output of tasks! This should be used very carefully.

***Quick testing runs ***

• debug only: run locally on debug dataset (defined in Processor) only
• explorative only: run on one file from each dataset using Dask

1. CoffeaProcessor (Histogramer or Exporter):

law run CoffeaProcessor --version test_01 --recipe selection --processor Processor --analysis-choice vh --year 2017 --debug

2. GroupCoffeaProcesses (after Histogramer): 3. Rebin: 4. Plotting or DatacardProducer:

law run PlotProducer --version test_01 --recipe selection --analysis-choice vh --log-scale --process-group plotting --model vh-model

law run DatacardProducer --version test_01 --recipe selection --local-scheduler --explorative

Multiclassification

First run data-preparation on free machein (vispa-gpu10 for example)

for SPLIT in 0 1 2 3 4; do law run vh.MulticlassDataprep --version dev1 --local-scheduler --split $SPLIT; done

law run vh.MulticlassStitched --version dev1 --local-scheduler --vh.MulticlassTraining-comet --vh.MulticlassTraining-htcondor-logs --workers 2

The workers 2 defines how many htcondor-tasks are submitted.

Fit - Combine

Combine

https://github.com/cms-analysis/CombineHarvester

Combine works with a workspace that is created from your datacard

Text

text2workspace.py </path/to/datacard.txt> -o workspace.root -m 125

Because the tool is old, you have to set the parameter for the higgs mass 125 (-m 125). This is not actually used but rather just a technical thing.

Upper limit

combine -M AsymptoticLimits workspace.root --verbose 1 --mass 125 --run expected --noFitAsimov -t -1

Multidim-fit

combine -M MultiDimFit workspace.root --verbose 1 --mass 125 -t -1 --robustFit 1  --setParameters r=1 --freezeParameters r

Datacard

Processes:

• Signal process: index
  \leq
  0
• Background process: index
  >
  0

Physics model:

• expr( proc A / proc B, scaling signal strength )