pkufudawei / boohft-calib

A tool for data-MC calibration of the "boosted heavy flavour jet tagger" in CMS

Tool for boosted heavy flavour jet tagger calibration

boohft-calib is a tool that serves for the data-MC calibration of the "boosted heavy flavour jet tagger" in CMS, based on the sfBDT coastline method. The tool runs under the Run 2 UL condition using NanoAODv9. It is designed for the calibration of any Xbb/Xcc type taggers composed of the branches in NanoAODv9.

Users should specify in a data card the tagger expression, pre-defined WPs, etc., and a signal ROOT tree for extraction of the necessary signal tagger shape. See details in the example YMAL card for calibrating the ParticleNet XbbVsQCD score.

The introduction of the method can be found in the BTV slides. Detailed documentation is provided in AN-21-005 (the sfBDT method).

The calibration results and all final & intermediate plots are showcased on the webpage, automatically generated after running a routine piloted by a YAML card. To see the example of the generated webpage, please refer to the above BTV slides.

Run the tool

1. Run on a local cluster

First set up the environment. We recommand to use Miniconda:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh -b -p ./miniconda  # for test: put the miniconda folder here
source miniconda/bin/activate
# clone the repo
git clone https://github.com/colizz/boohft-calib.git && cd boohft-calib
# install packages
conda env create -f conda_env.yml
conda activate boohft-calib

Run the tool in one command, e.g.,

python launcher.py cards/example_bb_PNetXbbVsQCD.yml

Try python launcher.py --help for more information on the command arguments.

Note: the tool uses 8 concurrent workers by default. On lxplus it will run by estimation 4 hrs for an entire routine. Sepcify more workers if you have more CPU resource.

2. Run on SWAN

Each routine will run by estimation 8 hrs on SWAN.

To run on SWAN, click the link, start a SWAN session with LCG96 Python3 stack (4 cores, 16GB), then open the launcher_swan.ipynb notebook and run all the blocks. This will launch a routine configured by the example card.

Configuration card

The configuration card (e.g., the example card cards/example_bb_PNetXbbVsQCD.yml) defines everything for a routine. As a brief summary, users should specify

• the type of calibration: can be bb, cc, or qq;
• the year of UL condition: can be 2016APV, 2016, 2017, or 2018;
• jet pT ranges for deriving separate SFs;
• the tagger information, including the tagger name/expression, the span, and the custom WPs defined in the user's analysis;
• info of a signal ROOT tree taken from the user's analysis which the tool uses for extracting the signal tagger shape.

See detailed explanation in the example card cards/example_bb_PNetXbbVsQCD.yml.

---

Update notes

v3.1.2 July 21, 2023

• Update: fix lumi uncertainty
• Update: apply no JERC correction to SV mass

v3.1.1 May 25, 2023

• Update: change the 20% frac_b/c/light variation in an overall manner (sync with mu-tagged method)
• Update: in case of a fit failure, enlarge the autoMCStats threshold and retry
• Feature: more text on plots to make it readable

v3.1.0 December 2, 2022

• Feature: add new uncertainties sources
• Feature: allow breaking down the full uncertainty list

v3.0.5 November 25, 2022

• Feature: allow using custom sfBDT models to replace the defalt one

v3.0.4 April 19, 2022

• Feature improved: allow expression to parse awkward-array indexing
• Reweight binning bug fix

v3.0.3 Mar 31, 2022

• Implement the qq calibration type

v3.0.2 Feb 5, 2022

• Implement the year condition for 2016APV and 2016

v3.0.1 Jan 29, 2022

• Support more command line arguments

v3.0.0 Jan 24, 2022

• Update the method to sfBDT coastline
• Update the framework to coffea (supports local run at present)

Previous version (till v2.1) developed in ParticleNet-CCTagCalib