mssm-htt-legacy-analysis

In this repository all software necessary for the MSSM HTT Legacy analysis starting from flat n-tuple level is stored. The software uses the ntuple_processor code included as submodule of the main repository. The software is written in python3 and uses RDataFrames.

Checkout instructions

The repository includes all necessary packages as submodules. Thus, a simple recursive clone of the repository should be sufficient.

git clone [email protected]:KIT-CMS/mssm-htt-legacy-analysis.git --recursive

Producing shapes

In order to run the shape production, the correct ROOT version needs to be sourced. For this a utility script is provided under utils/setup_root.sh. There is a top-level script for the shape production that allows to run the shape production locally or to write out a .pkl file containing a list of all created computational graphs for the different samples. Each of these created graphs can then be executed independently.

Shapes for control plots and GoF tests

The top-level script allows the creation of shapes for the analysis, control plots and GoF tests through the command line arguments --control-plots and --skip-systematic-variations. For the production of input shapes for GoF tests the --control-plots option is sufficient. For control plots including only statistical uncertainties the --skip-systematic-variations option needs to be set as well. An example command for the production of control plots (stat. only) in the tt channel for the 2017 data-taking period is:

source utils/setup_root.sh

python shapes/produce_shapes.py --channels tt --output-file control_shapes-2017-tt --directory /ceph/mburkart/Run2Legacy/ntuples/MSSM_Legacy/MSSM_2020_04_27_UseOfSMLegacy/2017/ntuples/ --tt-friend-directory /ceph/mburkart/Run2Legacy/ntuples/MSSM_Legacy/MSSM_2020_04_27_UseOfSMLegacy/2017/friends/{SVFit,FakeFactors}/ --era 2017 --num-processes 4 --num-threads 3 --optimization-level 1 --control-plots --skip-systematic-variations

Analysis shapes

Local production of shapes for control plots and GoF shapes for single channels is possible since they only need a small subset of the processes used in the analysis. For the analysis, this is not feasible since a large number of signal files needs to be processed as well. Therefore, a submit script for the analysis jobs is provided. Since the creation of the graphs takes quite some time it is not recommended to try to submit multiple channels and a large subset of the processes simultaneously. Thus, a split of the processes is provided in the submit script and even the submission of single processes is possible. The submit script will create the computational graphs locally and writes them to a file. Afterwards, it sets up the directories for the submission and the submission files and submits the jobs to the batch system. The submit files are set up to run on the ETP batch system without adaptations.

Before submitting the jobs it is recommended to create a directory on /ceph for the outputs and a second directory where the log files will be written to on the /work machine. These directories can then be symlinked into the repository structure via

ln -s /path/to/ceph/directory output
ln -s /path/to/work/directory log

To submit the shapes for the 2017 data-taking period the recommended way to invoke the script is:

for CHANNEL in et mt tt em
do
  for PROCESSES in backgrounds,sm_signals mssm_ggh_split{1..3} mssm_bbh_split{1..2}
  do
    bash submit/submit_shape_production.sh 2017 $CHANNEL $PROCESSES singlegraph $(date +%Y_%m_%d)
  done
done

This will create single core jobs for all pipelines except the nominal ones which will submitted as multicore jobs running the computational graph multithreaded on 8 cores.

The outputs of the jobs will be written to output/shapes/analysis_unit_graphs-2017-$CHANNEL-<processes>/output-single_graph_job-analysis_unit_graphs-2017-$CHANNEL-<processes>-<graph_number>.root where <processes> is the expanded subset of graphs of the submitted jobs and <graph_number> the number of the processed graph. The outputs can then be merged using the hadd command and further processed to obtain the input shapes for combine.

The provided utility script submit/merge_outputs.sh can be used to merge the output ntuples. Beforehand, the script submit/check_outputs.sh can be used to check if all expected output files have been written. Both scripts are invoked in the same way and expect the data-taking period, a comma-separated list of channels to be checked/merged and the tag used during the production of the ntuples as command line arguments.

The further processing of the merged ntuples can be done with two additional scripts. The first one estimates the shapes of derived processes like the fake factor method and the QCD multijet background from the created histograms and the second script converts the shapes to the format expected by combine.

bash shapes/do_estimations.sh $ERA $INPUT
bash shapes/convert_to_synced_shapes.sh $ERA $CHANNEL $VARIABLE $TAG

Here, the VARIABLE argument refers to the final discriminator the shapes should be converted for. This, allows to produce the input shapes for the SM and MSSM categories that are porduced with different discriminators. These shapes can then be used as input to the analysis repository. The ntuple_processor branch of the analysis repository should be used.

General structure

The configuration scripts are located in the config/shapes/ directory. The baseline selection is implemented in the channel_selection.py script. The input files are specified in the file_names.py script. The weights used for the different processes are set up in the process_selection.py script and the systematic variations are defined in the variations.py script.

Remarks on the structure of the ntuple-processor submodule

ToDo