DCTR: Pythia e+e- -> Z -> Dijets Datasets for Neural Network Reweighting

A collection of datasets for Neural Networks for Full Phase-space Reweighting and Parameter Tuning, generated with the Pythia 8.230 event generator. Particle-level e+e- -> Z -> dijet events with about 100 particles per event are clustered into jets using the anti-kt algorithm. Each jet is presented as a list of constituents with parameters for tuning, and the datasets include both training and test sets with specific observables.

Use Cases

• Training neural networks for full phase-space reweighting based on jet constituent lists.
• Parameter tuning of Pythia event generator settings based on the provided theta parameters (TimeShower:alphaSvalue, StringZ:aLund, StringFlav:probStoUD).
• Evaluating jet substructure using test set observables like Nsubjettiness (tau21, tau32) and Energy Correlation Functions (ECF_N3_B4, ECF_N4_B4).
• Comparing jet properties generated with default versus non-default Pythia parameters as indicated by the training labels (Y=0 or Y=1).

Strengths

• Datasets were generated with a specific, documented event generator (Pythia 8.230) and clustering algorithm (anti-kt with Fastjet 3.0.3).
• Training data includes clear labeling for default and non-default parameter jets, with parameter ranges explicitly defined (e.g., TimeShower:alphaSvalue in [0.10, 0.18]).
• Test datasets include multiple jet substructure observables (multiplicity, tau21, tau32, ECF_N3_B4, ECF_N4_B4) for evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred from the description after download.
• Row count is unknown, which may limit suitability assessment for large-scale model training.
• Last update date is unknown; freshness unverified.

Provenance

Source

Lawrence Berkeley National Laboratory (Author: Anders Andreassen), hosted on paperswithcode.

Collection Method

Simulated using the Pythia 8.230 event generator for e+e- -> Z -> dijet events, with jets clustered via the anti-kt algorithm.