# hls4ml LHC Jet Dataset (150 Particles): Five-Class Jet Tagging for FPGA Trigger Studies

## Facts

- Provider: Maurizio Pierini, Javier Duarte, Nhan Tran, Marat Freytsis (hls4ml / Fast Machine Learning Lab)
- License: CC-BY-4.0 (https://zenodo.org/records/3602260)
- Size: 18 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/fastmachinelearning/hls4ml_lhc_jets_150p
- Access: public, readable without an account
- DOI: https://doi.org/10.5281/zenodo.3602260
- Category: Physics
- Source: https://doi.org/10.1088/1748-0221/13/07/P07027
- Page: https://scigantic.com/archive/105e9f12-021a-531a-85fd-f0b14ef7df4b
- Keywords: Physics, Parquet, Hugging Face

## About

Simulated high transverse momentum (about 1 TeV) jets from LHC proton-proton collisions, labelled as gluon, light quark, W boson, Z boson or top quark, prepared by Pierini, Duarte, Tran and Freytsis for the hls4ml studies of low-latency neural network inference on FPGAs in the Level-1 trigger. Each jet keeps up to 150 constituents with 16 features (px, py, pz, energy, relative and rotated pT, eta and phi, delta-R, cos theta) and 53 jet-level substructure observables: N-subjettiness tau ratios, energy correlation functions C2, D2, M2 and N2, groomed masses (trimmed, mMDT, pruned, soft drop) and multiplicity. This repository, published by the fastmachinelearning organisation, reshapes the original HDF5 files to one Parquet row per jet with an integer label column, dropping only the jet images: 496,000 training and 124,000 validation jets cut from the Zenodo train archive and 260,000 test jets from its val archive, roughly balanced across the five classes. Files are data/{train,validation,test}-*.parquet (about 6 GB), loadable with Hugging Face datasets; a Hydra-driven loader that selects leading constituents, normalises features and returns torch tensors is included. The original record (10.5281/zenodo.3602260) is CC-BY-4.0; cite Duarte et al., JINST 13 P07027 (2018).

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("fastmachinelearning/hls4ml_lhc_jets_150p")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `data/test-00000-of-00002.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/105e9f12-021a-531a-85fd-f0b14ef7df4b
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/105e9f12-021a-531a-85fd-f0b14ef7df4b
- Over MCP: `get_schema_card` with id `105e9f12-021a-531a-85fd-f0b14ef7df4b` at https://api.scigantic.com/mcp

## Related topics

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