README.md

GNN Cleanup

GNN Cleanup is a tool designed to improve tracking performance in the Belle II experiment by filtering background hits in the Central Drift Chamber (CDC) using a Graph Neural Network (GNN)-based approach. By leveraging graph-based inference, the HitCleanupModule efficiently distinguishes between signal and background hits, enhancing tracking accuracy and computational efficiency, especially in high-background conditions.

Features

• Hit Cleanup Module: Filters background hits in the CDC using GNN inference, improving robustness to high-luminosity environments.

  • get CDC Hit Info Module: Extracts CDC information on hit level.
  • Graph Construction: Builds graph representations of CDC hits, including nodes (e.g., wire positions) and edges (e.g., spatial-temporal relationships).
  • Preprocessing: Applies feature clipping, scaling, and shifting to prepare graph data.
  • GNN-Based Filtering: Uses pretrained GNN models to perform hit filtering.

• Example Scripts:

  • A ready-to-use script demonstrates the full pipeline, including cleanup and tracking evaluation.
  • Runs full tracking workflows, including track finding and fitting, with or without the cleanup step.
  • Outputs tracking metrics, such as fitting charge efficiency, fake rate, and clone rate.

Setup and Installation

1. Clone the Repository:

   git clone git@gitlab.etp.kit.edu:grheine/gnn_cleanup.git
   cd gnn_cleanup

2. Run Setup Script: Open your terminal and run:

   source setup.sh

   This sets up basf2 release 6, sources an virtual python environment and sets the PATH variables.

Example Usage

To test the gnn_cleanup, run the example script with the appropriate options:

python3 examples/testing_withoutb2luigi.py --n 100

Using other event files

For using other events first update the testing_config.json file. There you can specify the following settings:

• load_path: path to existing event root file
• expList: experiment number list for EventInfoSetter
• runList: run number list for EventInfoSetter
• bg_file_list_path: path to background file lists

You can either use existing event root files by setting the load_path to point to your files or generate new MC events by using the --generate flag and updating the bg_file_list_path, expList and runList.