JetPointNet

This repository contains the code used to develop particle-flow models based on pointcloud data for jets data.

Installation

The current version has been tested with the following environment settings:

python 3.10.14
awkward 2.6.4
uproot 5.3.7
numpy 1.26.4
pandas 2.2.2
tensorflow 2.16.1
keras 3.3.3

For a minimal installation you can first create a conda environment with conda create --name pointcloud python=3.10. Then make use of the requirements.txt file and run:

conda activate pointcloud
pip install -r requirements.txt # --no-cache-dir

Please note that --no-cache-dir option is suggested.

Usage

Data Preprocessing:

Note: preprocessing with the ttbar set is possible, however with larger sets you should attempt to use already processed npz files to avoid a lengthy re-processing. Additionally for other sets you should use a grid job for the processing (see last set of instructions). The current data set is available as npz file on CERNBox, if you are using them you can skip these steps.

• update configurations and paths inside python_scripts/data_processing/preprocessing_header.py
• run python_scripts/data_processing/jets_root_to_awk.py to convert root data into awkward array format. This reads data from FILE_LOC or FILES_DIR and GEO_LOC and saves results into the AWK_SAVE_LOC (all defined in preprocessing_header.py)
• run python_scripts/data_processing/jets_awk_to_npz.py to convert awkward array data into npz format. This reads data from AWK_SAVE_LOC

Model Training:

• once you have preprocessed data, then you can start playing with training scripts at python_scripts/jets_training/. In particular, the script jets_train.py is used for running individual models and jets_sweep.py is used to preform sweeps.

Notes

This branch assumes the repository is located at HOME_PATH / workspace/jetpointnet. Also, raw data files are read from eos (namely /eos/home-m/mswiatlo/forLuca/ and /eos/home-m/mswiatlo/images/truthPerCell), while derived files are stored inside the repo at pnet_data/ folder and organised in subfolders.

Preprocessing dependency tree

Known Issues

• The early stopping of the train loop can cause the job to crash when enabled