# Pythia 8 Quark and Gluon Jets (Komiske, Metodiev, Thaler), Parquet with Particle Transformer Features

## Facts

- Provider: Patrick Komiske, Eric Metodiev, Jesse Thaler (MIT); Parquet conversion by jet-universe
- License: CC-BY-4.0 (https://zenodo.org/records/3164691)
- Size: 12 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/jet-universe/quark_gluon
- Access: public, readable without an account
- DOI: https://doi.org/10.5281/zenodo.3164691
- Category: Physics
- Source: https://arxiv.org/abs/1810.05165
- Page: https://scigantic.com/archive/f3f3ae40-6e5b-5dc7-9422-a023722adadf
- Keywords: Physics, benchmark, Parquet, Hugging Face

## About

Two million simulated quark and gluon jets for quark/gluon discrimination, the standard benchmark introduced with Energy Flow Networks (Komiske, Metodiev and Thaler, JHEP 2019) and used by ParticleNet and Particle Transformer. Jets come from Pythia 8.226 at sqrt(s)=14 TeV: quark jets from qg to Z+q and gluon jets from qqbar to Z+g with the Z decaying to neutrinos, clustered with FastJet anti-kT R=0.4 and selected at jet pT 500 to 550 GeV and |y|<1.7, with equal numbers of quark and gluon jets and no detector simulation. This copy, maintained by the jet-universe group behind JetClass, converts the original NumPy archives into Parquet with the Particle Transformer feature schema: per-jet pT, eta, phi, energy, mass and constituent count, and per-constituent px, py, pz, energy, delta-eta and delta-phi to the jet axis, PDG id, charge and particle-type flags (charged hadron, neutral hadron, photon, electron, muon). Splits are 1,799,982 training and 199,998 test jets in data/train-0000N-of-00009.parquet and data/test-00000-of-00001.parquet (about 3 GB); load with Hugging Face datasets, pandas or awkward. The upstream Zenodo record (10.5281/zenodo.3164691) is CC-BY-4.0; cite Komiske, Metodiev and Thaler, arXiv:1810.05165.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("jet-universe/quark_gluon")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/test-00000-of-00001.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/f3f3ae40-6e5b-5dc7-9422-a023722adadf
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/f3f3ae40-6e5b-5dc7-9422-a023722adadf
- Over MCP: `get_schema_card` with id `f3f3ae40-6e5b-5dc7-9422-a023722adadf` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Benchmark datasets](https://scigantic.com/topics/benchmark)
- [Physics datasets](https://scigantic.com/topics/physics)
