Top Quark Tagging Reference Dataset (Parquet)

Provider
Gregor Kasieczka, Tilman Plehn, Jennifer Thompson, Michael Russel (original); Parquet conversion by dl4phys
License
CC-BY-4.0 (source)
Size
7 files
Format
Parquet (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/dl4phys/top_tagging)
Access
Free and public. No account with the data provider is needed.
DOI
10.5281/zenodo.2603256
Category
Physics
Source
https://arxiv.org/abs/1902.09914

About Top Quark Tagging Reference Dataset (Parquet)

The Top Quark Tagging Reference Dataset of Kasieczka, Plehn and colleagues, the community benchmark behind the 2019 comparison paper 'The Machine Learning Landscape of Top Taggers' (SciPost Physics). Monte Carlo events from 14 TeV proton-proton collisions were generated with Pythia8 and the Delphes ATLAS detector card, without multiple interactions or pile-up, clustered into anti-kT R=0.8 jets with pT 550 to 650 GeV; signal jets are hadronic tops and background is QCD. Each row has the four-momenta (E, px, py, pz) of the leading 200 jet constituents sorted by transverse momentum and zero-padded, the truth top four-momentum, and the label is_signal_new. This Parquet conversion, made for the Hugging Face 'Deep Learning for Particle Physicists' course, keeps the original splits: 1,211,000 train, 403,000 validation and 404,000 test jets; files are train.parquet, validation.parquet and test.parquet, about 2.3 GB in total (measured 2026-11-14). Load with datasets, pandas or pyarrow. The original Zenodo release (10.5281/zenodo.2603256) is CC-BY-4.0; cite Kasieczka et al. 2019.

How to load Top Quark Tagging Reference Dataset (Parquet) in Python

from datasets import load_dataset
ds = load_dataset("dl4phys/top_tagging")
print(ds)

Data structure

File types in sampled files: Parquet

Files

Example file: test.parquet

Analyze this dataset without downloading it

On Scigantic, Top Quark Tagging Reference Dataset (Parquet) opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Physics datasets