# JetClass-II: 188-Class Large-Radius Jet Dataset for Jet Foundation Models (LHC Delphes Simulation)

## Facts

- Provider: Congqiao Li, Huilin Qu et al. (jet-universe, CERN and Peking University)
- License: MIT (https://huggingface.co/datasets/jet-universe/jetclass2)
- Size: 2,013 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/jet-universe/jetclass2
- Access: public, readable without an account
- Category: Physics
- Source: https://arxiv.org/abs/2405.12972
- Page: https://scigantic.com/archive/d6b03ce4-2c59-50e4-8295-a44b9a9976e8
- Keywords: Physics, Parquet, Hugging Face

## About

JetClass-II is a simulated dataset of large-radius (anti-kT R=0.8) jets for training jet tagging and jet foundation models for physics at the Large Hadron Collider, released by Huilin Qu and collaborators with the Sophon model (arXiv:2405.12972). Events are generated with MadGraph5_aMC@NLO and Pythia 8 and passed through Delphes with a CMS-like detector, covering a wide range of jet pT and mass. Jets fall into three parts: Res2P (generic X to 2-prong resonances, 20M training jets), Res34P (X to 3- or 4-prong resonances via Y1Y2, 86M) and QCD multijet background (28M), split 4:1:1 into train, validation and test, with a jet_label over 188 fine-grained classes by initiating partons and leptons. Each row holds reconstructed particle-flow constituents (four-momenta, impact parameters d0 and dz with errors, charge and particle-ID flags), jet features (soft-drop mass, N-subjettiness tau1 to tau4), the matched generator-level jet and its constituents with production vertices, and truth resonance and decay-product kinematics. Data are Parquet files of 100k jets each (Res2P_0000.parquet and so on, about 740 GB in total); read them with pyarrow, awkward or pandas, or with the weaver and Sophon code. Released under the MIT licence by the publishers; cite Li, Qu et al., arXiv:2405.12972.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("jet-universe/jetclass2")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/QCD_0000.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/d6b03ce4-2c59-50e4-8295-a44b9a9976e8
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/d6b03ce4-2c59-50e4-8295-a44b9a9976e8
- Over MCP: `get_schema_card` with id `d6b03ce4-2c59-50e4-8295-a44b9a9976e8` at https://api.scigantic.com/mcp

## Related topics

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