PIDNet

Deep Learning based 3D Particle Identification

Code, dataset, and trained models for the paper "PIDNet: Deep Learning based $\bold{e^-/\pi^0}$ separation in the Single Phase Liquid Argon TPC"

If you find this code useful in you work, please cite:

@article{bay2020pidnet,
  title={PIDNet: Deep Learning based $\bold{e^-/\pi^0}$ separation in the Single Phase Liquid Argon TPC},
  author={},
  journal={arXiv preprint arXiv:},
  year={2020}
}

Particle Topologies

Getting Started

Installation

• Clone this repo:

cd ~
git clone https://https://github.com/yasinalm/PIDNet
cd PIDNet

Prerequisites

• Linux (Tested on Ubuntu 16.04)

• NVIDIA GPU (Tested on Nvidia GTX 1080 Ti)

• CUDA, CuDNN

• Python 3

• tensorflow-gpu>=2.0.0

• Tensorboard>=2.0.0

• scipy

• Install Tensorflow 2.0.0+ and other dependencies.

  • For pip users, please type the command pip install -r requirements.txt.
  • For Conda users, you can use an installation script ./scripts/conda_deps.sh. Alternatively, you can create a new Conda environment using conda env create -f environment.yml.

Dataset

• Our dataset consists of particles generated in Monte Carlo simulations.
• Place your dataset in ./dataset folder and use the codes in ./prepapre_data folder to pre-process.

Dataset Organization

Data needs to be arranged in the following format:

.
├── dataset5d
│   └── 1500sp_evts
│       ├── electron-38323
│       ├── muon-62190
│       ├── pionminus-39144
│       ├── pionzero-35674
│       ├── proton-36793
│       ├── test_files35k_5p.csv
│       └── train_files35k_5p.csv
└── PIDNet
    ├── figs
    │   └── confusion_matrix_pointnet.png
    ├── logs
    │   └── checkpoints
    │       ├── checkpoint
    │       ├── iter-102828.data-00000-of-00002
    │       ├── iter-102828.data-00001-of-00002
    │       └── iter-102828.index
    ├── src
    │   ├── dataset_utils.py
    │   ├── eval_test_set.py
    │   ├── eval_test_set_tsne.ipynb
    │   ├── inference.py
    │   ├── match_pred_momentum.ipynb
    │   ├── model.py
    │   ├── plot_pred_momentum.ipynb
    │   ├── roc_curve.ipynb
    │   ├── train.py
    └── utils
        ├── helpers.py
        └── visualize.py

Network Architecture

Steps to reproduce the results

1. Prepapare the data using the command: 'prepare_data/python prepare_fixedsp_data.py'
2. Create a list of train and test files with the Jupyter notebook file prepare_data/generate_train_test_files.ipynb
   1. Just sequentially run the code blocks in the notebook.
3. Run the model in the pid_net folder: 'python src/train.py'

Training

To train a model:

python -u src/train.py |& tee logs/log_training_$(date '+%Y%m%d-%H%M%S').txt

To train with custom parameters:

python -u src/train.py --n_classes=5 --batch_size=64 --epochs=50  |& tee logs/log_training_$(date '+%Y%m%d-%H%M%S').txt

To resume training from a checkpoint (if directory, uses the latest checkpoint in the directory):

python -u src/train.py --init_weight=${ckpt_dir}  |& tee logs/log_training_$(date '+%Y%m%d-%H%M%S').txt

If you have multiple GPUs on your machine, you can also run the multi-GPU version training:

CUDA_VISIBLE_DEVICES=0,1 python -u src/train.py |& tee logs/log_training_$(date '+%Y%m%d-%H%M%S').txt

• To see more intermediate results, check out ./logs/checkpoints/.
• To view training results and loss plots, run tensorboard --logdir logs/ and click the URL http://localhost:6006.

Pre-trained Models

• The pretrained model is saved at ./logs/checkpoints/.

Testing

To test the model:

python -u src/eval_test_set.py --init_weight=${ckpt_dir} |& tee logs/log_test.txt

• The statistics will be prompted to screen and confusion matrix result will be saved to a png file here: ./figs/confusion_matrix.png.

Confusion Matrix, ROC Curve, and T-SNE Visualization

License

This project is licensed under the MIT License - see the LICENSE file for details

Reference

If you find our work useful in your research please consider citing our paper:

@article{bay2020pidnet,
  title={PIDNet: Deep Learning based $\bold{e^-/\pi^0}$ separation in the Single Phase Liquid Argon TPC},
  author={},
  journal={arXiv preprint arXiv:},
  year={2020}
}