NOTE: This README does not contain commands needed to run newer scripts as of August 2021. For the updated files, please refer to the appendices of our report.
Setup the environment
virtualenv env-ml
source env-ml/bin/activate
pip install tensorflow==2.0.0-beta1
pip install sklearn
pip install pandas
pip install ray
Install ROOT from src https://root.cern.ch/downloading-root
Let's assume ROOT is installed under $HOME/local/root
Clone repository
git clone https://github.com/IMFardz/AngryTops.git
module load cuda/9.0.176
module load cudnn
source $HOME/env-ml/bin/activate
source $HOME/local/root/bin/thisroot.sh
AngryTops is a pipeline for training, testing and evaluating neural networks designed to compute the kinematic reconstruction of ttbar events formed from pp collisions. This package is divided into five sub-package.
EventGeneration: Takes ttbar events produced from the Monto Carlo pipeline (MadGraph8 + Pythia5 + Delphes) and converts them into a csv file format. The main script, root2csv.py, selects events and specific branches from the ROOT Trees, applies cuts and writes event info, input columns and output columns into a csv format. Several helper scripts that are used here are tree_traversal.py, helper_functions.py and data_augmentation.py. HistCSV.py is used to output the distributions of the events stored in the csv files.
ModelTraining: The main pipeline for creating, training and saving neural networks. Different network architectures are written in models.py and cnn.py. The training and testing data is withdrawn from the csv files in FormatInputOutput.py. The model is trained, loaded and saved in train_simple_model.py.
Plotting: Pipeline for producing the distributions of the true test values and the fitted values from the network. The scripts are run in the order: fit.py -> histograms.py -> plot_observables.py. For models that output only a single distribution, use plot_single.py should be used instead.
HyperParameterSearch: Pipeline for searching hyper-parameter spaces. Model architectures are defined in test_models.py and their corresponding hyper-parameter spaces are in param_spaces.py. HyperOptSearch.py is the main script used for this pipeline.
PostTraining_Analysis: Routine for training a model and then retraining the model on a subset of events, determined
CheckPoints: Where trained models are saved. archive: Contains old scripts AtlasDocs: Used for formatting plots with ROOT ray_results: Contains results from hyperparameter searches combine_images.py: Used to create comparison plots between models.