LEE_calculation

The code is adapted to be able to use for the HPlus analysis.

Toy Generation for Hplus ->cb analysis

Requirements

• ROOT 6.22.06
• Python 3 (see requirements.txt for additional packages)

Step 0: Preparation

1. Create a working directory
2. Get the workspaces and workspace building log files
3. Get ntuples for hadhad, lephad SLT + LTT

## Step 1: Discriminant Binning

Parse the WSMaker workspace building log files to get the bin edges
used for the final discriminant.

```bash
makepickles.py (for our studies)

Six pickle files should be produced that store the bin edges indices used for rebinning -> 6 files -> 6 categories

Step 2: Create Asimov Datasets

Produce histograms of the Asimov dataset providing:

1. The background expectation
2. The effective number of MC events (tau) for the Barlow-Beeston method

mkdir asimov

for mass in 60 70 80 90 100 110 120 130 140 150 160; do
    python3 scripts/makeAsimov_hplus.py data/workspaces/${mass}.root -m ${mass} -o asimov/asimov_${mass}.root
done

(
    cd asimov
    [[ -e asimov_merged.root ]] && rm asimov_merged.root
    hadd asimov_merged.root asimov_*.root
)

The histograms will store the value of the observables obs_*, effective MC statistics tau_* for all channels: {4jex3bex,4jex4bex,5jex3bex,5jex4bex,6jex3bex,6jex4bex,} and for all discriminants _mass*.

Step 3: Make Dataframes

Ntuples are converted to pandas dataframes for ease of use during toy generation. Additionally, the MVA scores are already discretized according to the binning used in the final fit.

python3 scripts/makeHPlusDataframes_4j3b.py data/datainputs.txt -o dataframes/dataframe_4j3b.h5
python3 scripts/makeHPlusDataframes_4j4b.py data/datainputs.txt -o dataframes/dataframe_4j4b.h5
python3 scripts/makeHPlusDataframes_5j3b.py data/datainputs.txt -o dataframes/dataframe_5j3b.h5
python3 scripts/makeHPlusDataframes_5j4b.py data/datainputs.txt -o dataframes/dataframe_5j4b.h5
python3 scripts/makeHPlusDataframes_6j3b.py data/datainputs.txt -o dataframes/dataframe_6j3b.h5
python3 scripts/makeHPlusDataframes_6j4b.py data/datainputs.txt -o dataframes/dataframe_6j4b.h5

Step 4: Expected Correlation Between Bins in Data

Estimates the expected correlation matrix of data yields per bin using a bivariate Poisson model.

mkdir correlation_matrices

python3 scripts/makeCorr_hplus.py dataframes/dataframe_4j3b.h5 -o correlation_matrices/corr_4j3b.h5
python3 scripts/makeCorr_hplus.py dataframes/dataframe_4j4b.h5 -o correlation_matrices/corr_4j4b.h5
python3 scripts/makeCorr_hplus.py dataframes/dataframe_5j3b.h5 -o correlation_matrices/corr_5j3b.h5
python3 scripts/makeCorr_hplus.py dataframes/dataframe_5j4b.h5 -o correlation_matrices/corr_5j4b.h5
python3 scripts/makeCorr_hplus.py dataframes/dataframe_6j3b.h5 -o correlation_matrices/corr_6j3b.h5
python3 scripts/makeCorr_hplus.py dataframes/dataframe_6j4b.h5 -o correlation_matrices/corr_6j4b.h5

Step 5: Generate Poisson RVS

This generates correlated Poisson random variables for the observables in the 6 category regions channels:

mkdir poisson_rvs

python3 scripts/generateFromCorr_hplus.py correlation_matrices/corr_4j3b.h5 asimov/asimov_merged.root \
    -c 4j3b -o poisson_rvs/rvs_4j3b.h5

python3 scripts/generateFromCorr_hplus.py correlation_matrices/corr_4j4b.h5 asimov/asimov_merged.root \
    -c 4j4b -o poisson_rvs/rvs_4j4b.h5

python3 scripts/generateFromCorr_hplus.py correlation_matrices/corr_5j3b.h5 asimov/asimov_merged.root \
    -c 5j3b -o poisson_rvs/rvs_5j3b.h5

    ... 

Step 6: Generate Global Observables (Barlow-Beeston)

This will create multiple root-files containing trees with the values of the global observables.

mkdir gamma_globs

makeGammaGlobsToys_hplus.py dataframes/dataframe_4j3b.h5 asimov/asimov_merged.root \
    -o gamma_globs -c 4j3b

... ( all regions)

Step 7: Generate Global Observables (Others)

All other (Gaussian-constrained) global observables are treated as fully correlated. The script takes all workspaces so that it can figure what the names of all NPs are.

mkdir other_globs

python3 scripts/makeAlphaGlobsToys_hplus.py \
 data/workspaces/{60,70,80,90,100,110,120,130,140,150,160}.root \
 -o other_globs/alphas.root

Step 9: Build Workspace Inputs

This (mostly technical) step translates the pseudo-data that was generated in the final fit binning to the initial histogram binning so that it can be used in the workspace building as if it were real data.

Pseudo-data:

mkdir ws_inputs

makePseudoDataHists_hplus.py poisson_rvs/rvs_4j3b.h5 -c 4j3b -o ws_inputs/
...

Global observables:

The global observables are stored as trees where the index of the entry corresponds to the toy experiment. We can just merge the ROOT files for all global observables:

for mass in 60 70 80 90  \
            100 110 120 130 140 150 160; do
         
    hadd ws_inputs/toy_globs_${mass}.root \
        gamma_globs/toy_globs_{4j3b,4j4b,5j3b,5j4b,6j3b,6j4b}_${mass}.root \
        other_globs/alphas.root
done