RS3L: A jet tagging dataset for self-supervised learning based on re-simulation

Jet tagging dataset used for "Re-Simulation-based Self-Supervised Learning" (RS3L, arXiv:2403.07066). Simulated partons are re-showered with various parton shower configurations and reconstructed with the Delphes3 detector software. The singletons array contains information about the jet and has dimensions N_examples x N_augmentations x N_singletons. The augmenations are ordered by: 1. nominal scenario: jet showered with Pythia82. changing the numerical seed in Pythia83. changing the scale controlling the probability for final state radiation by 1/sqrt(2)4. changing the scale controlling the probability for final state radiation by sqrt(2)5. using Herwig7 as parton shower

The variables stored in the singletons array are:

singletons = ['jettype','parton1_pt','parton1_eta','parton1_phi','parton1_e','parton2_pt','parton2_eta','parton2_phi','parton2_e', 'jet_pt','jet_eta','jet_phi','jet_e','jet_msd','jet_n2']

For gluon-initiated jets, parton1 and parton2 are the two gluon daughters (see below). For any other jet type, parton1 is the main particle (W, H, Z, or single quark), and parton2 is filled with 0s.

The jet_pflow_cands contains the particles clustered into the jet. The array has a shape N_examples x N_augmentations x N_features x N_particles.The features per particle are:

features = ['pt','relpt','eta','phi','dr','e','rele','charge','pdgid','d0','dz']The jettype gives the parton at the origin of the jet: jettype: 1: q2: c3: b4: H->bb5: g->qq6: g->cc7: g->bb8: g->gg9: W->two quarks10: Z->qq11: Z->bb