Repository for my SURF Caltech-CMS project on deep learning for imaging calorimetry in the LCD (Linear Collider Detector) at CERN.
This work was published in the Deep Learning in Particle Physics (DLPS) workshop at NIPS 2017.
B. Hooberman, V. Barin Pacela, M. Zhang, W. Wei, G. Khattak, S. Vallecorsa, A. Farbin, J-R. Vlimant, F. Carminati, M. Spiropulu, M. Pierini. Calorimetry with Deep Learning: Particle Classification, Energy Regression, and Simulation for High-Energy Physics. DLPS 2017, NIPS 2017, Long Beach, CA, USA.
This repository contains data pre-processing functions (src/preprocessing.py), model topologies (src/model.py) and post-processing functions (src/analysis.py) to analyze model performance.
You can find example notebooks in examples. Older notebooks (not updated) can be found in the NotebooksLCD repository.
To submit a training job to a gpu, modify src/train.py adapting it to your data.
To get predictions of the test set, use src/test.py.
You can read the documentation in docs/Documentation.md. UML activity diagrams are in the same directory.
After cloning this repository and entering it, execute pip install . --user.
Notice that there are inconsistencies with how tensorflow should be installed.
Code in Python 2.7.
Pre-processing dependencies: Danny Weitecamp's CMS_Deep_Learning package for the data generator.
Machine Learning dependencies: Keras 1.2.2 with Tensorflow 0.12.0 in the backend.
Post-processing dependencies: Matplotlib 1.4.3, Scipy 1.0.0.