johannbrehmer / higgs_inference

Code repository for the paper "Constraining Effective Field Theories with Machine Learning"

Code repository for the paper "Constraining Effective Field Theories with Machine Learning"

Johann Brehmer, Kyle Cranmer, Gilles Louppe, and Juan Pavez

Note: If you want to use these methods for a particle physics problem, please have a look at our new tool MadMiner -- you will likely find that much easier to use than this repo!

Folder structure

• cluster_scripts: SLURM scripts that start the preprocessing and experiments on the NYU HPC cluster.
• data: The data set, including both the original weighted event sample as well as the unweighted training, calibration, and evaluation samples. Some of these are quite large and not on GitHub.
• evaluation: IPython notebooks that extract metrics and figures from the experiments.
• figures: Here the figures are stored.
• higgs_inference: The main folder for the inference experiments.
  • models: The keras model code at the heart of many inference strategies.
  • strategies: Training and evaluation routines for the different inference strategies.
  • various: Different utility functions.
  • experiments.py: Main executable that starts the different training and evaluation pieces.
  • settings.py: Most settings and constants, including the main directories, event numbers, architecture parameters, and benchmark thetas.
• preprocessing: Unweighting routines that generate the different training, calibration, and evaluation samples from the original weighted event file.
• postprocessing: Code for the Neyman construction.
• prototypes: Toy experiments and cross-checks on other data. Includes the flow folder with a PyTorch implementation of normalizing flows.
• results: The predictions of the different algorithms on the test data set.