How to Understand Limitations of Generative Networks - Generator datasets

These are the datasets used in "How to Understand Limitations of Generative Networks". The preprint is available on arXiv at: https://arxiv.org/abs/2305.16774

Four files are used in Sec.5 "Event generation":

• ev_truth.h5 is the true reconstruction-level sample used during training;
• ev_masspeak.h5 is the sample generated from the neural network of Sec. 5.1;
• ev_inn.h5 is the state-of-the-art sample used in Sec. 5.2;
• ev_binn.h5 collects the Bayesian samples of Sec. 5.3;

The remaining three files are used for the "Calorimeter simulation" section.
These are named according to the particle originating the shower: "calo_eplus.hdf5" for positrons, "calo_gamma.hdf5" for photons, and "calo_piplus.hdf5" for pions. Each sample contains 100k showers.