mcreweight 0.2.5

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 pyproject.toml

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 .

Usage

To run the reweighting:

run-reweight --path_data <path_to_data.root> \
    	     --path_mc <path_to_mc.root> \
    	     --vars <variable_list> \
    	     --monitoring_vars <monitoring_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> \
    	      --training_sample <training_sample> \
    	      --application_sample <application_sample> \
    	      --method <method_for_reweighter> \
    	      --monitoring_vars <monitoring_variable_list> \
    	      --output_path <output_file.root>

Options

For the reweighting (run-reweight):

Input files:

• --path_data: Path to the data control sample (required)
• --tree_data: Name of the tree in the data control sample (default: "DecayTree")
• --path_mc: Path to the MC control sample (required)
• --tree_mc: Name of the tree in the MC control sample (default: "DecayTree")
• --mcweights_name: Name of the branch for weights in the MC sample (default: None)
• --sweights_name: Name of the sweights column in the data (default: "sweight_sig")
• --mc_label: Label for the MC sample (default: "MC")
• --data_label: Label for the data sample (default: "Data")

Variables:

• --vars: List of variables to use for reweighting (default: ["B_DTF_Jpsi_P", "B_DTF_Jpsi_PT", "nLongTracks", "nPVs"])
• --monitoring_vars: List of variables to plot (default: None)

Reweighter configuration:

• --sample: Sample name for the dataset (default: "bd_jpsikst_ee")
• --trials: Number of trials for the gradient boosting reweighting (default: 10)
• --test_size: Proportion of the dataset to include in the test split (default: 0.3)
• --n_folds: Number of folds for k-folding reweighting (default: 4)
• --n_bins: Number of bins for binning reweighting (default: 20)
• --n_neighs: Number of nearest neighbors for binning reweighting (default: 3)

Output:

• --weightsdir: Directory to save weights (default: "weights")
• --plotdir: Directory to save plots (default: "plots")

Additional options can be found by running:

run-reweight --help

For the application of the weights (apply-weights):

Input files:

• --path_mc: Path to the MC signal sample (required)
• --tree_mc: Name of the tree in the MC signal sample (default: "DecayTree")
• --mcweights_name: Name of the branch for weights in the output ROOT file (default: None)
• --path_data: Path to the data sample for comparison (default: None)
• --tree_data: Name of the tree in the data sample (default: "DecayTree")
• --sweights_name: Name of the sweights column in the data (default: "sweight_sig")

Variables:

• --vars: List of variables to use for reweighting (default: ["B_DTF_Jpsi_P", "B_DTF_Jpsi_PT", "nLongTracks", "nPVs"])
• --training_vars: List of variables used for training (default: ["B_DTF_Jpsi_P", "B_DTF_Jpsi_PT", "nLongTracks", "nPVs"])
• --monitoring_vars: List of variables to plot (default: None)

Configuration:

• --training_sample: Sample name for the dataset (default: "bd_jpsikst_ee")
• --application_sample: Sample name for the application of weights (default: "bd_jpsikst_ee")
• --method: Method to apply weights (choices: "gbreweighter", "kfolding", "binning", default: "gbreweighter")
• --weightsdir: Directory to save weights (default: "weights")
• --plotdir: Directory to save plots (default: "plots")

Output:

• --output_path: Path to save the output ROOT file (required)
• --output_tree: Name of the tree in the output ROOT file (default: "DecayTree")

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 \
    	     --monitoring_vars B_ETA nFTClusters nVPClusters nEcalClusters \
    	     --sample bd_jpsikst_ee \
    	     --n_trials 5 \
    	     --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  \
    	      --training_sample bd_jpsikst_ee \
    	      --application_sample bd_jpsikst_ee \
    	      --method gbreweighter \
    	      --monitoring_vars B_ETA nFTClusters nVPClusters nEcalClusters \
    	      --output_path mc/signal_tuple_reweighted.root

Contact

For questions, please contact the repository maintainer.