This repository contains the source code used to produce the results of
"Unifying Simulation and Inference with Normalizing Flows" by Haoxing Du, Claudius Krause, Vinicius Mikuni, Benjamin Nachman, Ian Pang and David Shih, [arxiv: 2404.18992]
We consider a new sampling calorimeter (ECAL+HCAL) version of the toy detector used in the original CaloGAN. This new calorimeter setup includes a HCAL which was not included in the setup used the most recent CaloGAN update. The original dataset included energy contributions from both active and inactive calorimeter layers, whereas our new dataset only includes energy contributions from the active layers as would be available in practice. In our calorimeter setup, the sampling fractions for the ECAL and HCAL are
The new dataset can be found at https://zenodo.org/records/11073232.
Please see https://gitlab.com/claudius-krause/caloflow for instructions on training CaloFlow.
To use trained flows to compute likelihood for calibration task, run
python MLE_analysis-100k_densities.py --weights_dir =FLOW_WEIGHTS_DIR_NAME --results_dir=RESULTS_DIR_NAME