mcreweight 0.1.3

A package for reweighting MC samples to match data

mcreweight

mcreweight is a python library to perform Monte Carlo event reweighting based on multiplicity and kinematic variables. The tool is using GBReweighter, a classifier-based method implemented in hep_ml package, and it supports automated hyperparameter tuning with Optuna. A folding approach over the reweighter is also applied, and performances are compared with the ones from the bins reweighting.

[!WARNING] Bins reweighting works fine for one or two dimensional histograms, but it is unstable and inaccurate for higher dimenstions

Requirements

• Python 3.8+
• Required packages listed in requirements.txt

Setup

If you run in a lb-conda environment, as

lb-conda default

consider exporting the PATH to run scripts/run_reweighting.py and scripts/apply_weights.py with commands run-reweight and apply-weights. In this case, use

export PATH=$PATH:/path/to/.local/bin

Installation

From PyPl

pip install mcreweight

From Gitlab

git clone https://github.com/tfulghes/mcreweight.git
cd mcreweight
pip install -e .

You may need the following dependencies:

pip install -r requirements.txt

Usage

To run the reweighting:

run-reweight --path_data <path_to_data.root> \
    	     --path_mc <path_to_mc.root> \
    	     --vars <variable_list> \
    	     --sample <sample> \
    	     --trials <optuna_tests> \
    	     --test_size <test_sample_size>

To apply the weights to the signal MC:

apply-weights --path_mc <path_to_mc.root> \
    	      --vars <variable_list> \
    	      --sample <sample> \
    	      --method <method_for_reweighter>

Options

For the reweighting:

• --path_data: Path to the ROOT file containing real data events.
• --path_mc: Path to the ROOT file containing Monte Carlo events.
• --vars: List of variables to use for reweighting.
• --sample: Name of the sample to process (used for output naming and selection).
• --sweights_name: Name of sweights branch in the data
• --trials: Number of Optuna optimization trials for hyperparameter tuning.
• --test_size: Fraction of events to use as the test set (e.g., 0.3 for 30%).
• --n_folds: Number of folds for the k-folding reweighting
• --n_bins: Number of bins for binning reweighting
• --n_neighs: Number of nearest neighbors for binning reweighting

Additional options can be found by running:

run-reweight --help

For the application of the weights:

• --path_mc: Path to the MC ROOT file to which weights will be applied.
• --vars: List of variables in MC to use for reweighting.
• --training_vars: List of variables used for the training.
• --sample: Name of the sample to process.
• --method: Reweighting method to use (e.g., gbreweighter, kfolding or binning).

Additional options can be found by running:

apply-weights --help

Example

Reweighting:

run-reweight --path_data data/control_sample_tuple.root \
    	     --path_mc mc/control_sample_tuple.root \
    	     --vars B_DTF_Jpsi_P B_DTF_Jpsi_PT nLongTracks nPVs \
    	     --sample bd_jpsikst_ee \
    	     --trials 25 \
    	     --test_size 0.3 

Application of the weights:

apply_weights --path_mc mc/signal_tuple.root \
    	      --vars B_P B_PT nLongTracks nPVs  \
    	      --training_vars B_DTF_Jpsi_P B_DTF_Jpsi_PT nLongTracks nPVs  \
	      --sample bd_jpsikst_ee\
    	      --method gbreweighter

Contact

For questions, please contact the repository maintainer.