# CMS Open Data HH to 4b: Run 2 Collision Data, Hemisphere-Mixed Pseudo-Data and Simulation for Multijet Background Modelling

## Facts

- Provider: CMS Collaboration (CERN Open Data record cms-93965); Hugging Face packaging by Alejandro Gomez Espinosa
- License: CC0-1.0 (https://opendata.cern.ch/record/cms-93965)
- Size: 205 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/alefisico/LHC_HH4b_mixeddata
- Access: public, readable without an account
- DOI: https://doi.org/10.7483/OPENDATA.CMS.WWHU.D4ON
- Category: Physics
- Source: https://opendata.cern.ch/record/cms-93965
- Page: https://scigantic.com/archive/8c83bc81-bef0-5d63-833a-44d99db85bea
- Keywords: Physics, Parquet, Hugging Face

## About

Flat Parquet tables from the CMS Collaboration's CERN Open Data record cms-93965 (DOI 10.7483/OPENDATA.CMS.WWHU.D4ON), built for machine-learning background estimation in searches for Higgs boson pair production in the four-bottom-quark final state (HH to bbbb) with CMS Run 2 proton-proton collisions at 13 TeV (2016 to 2018). The release contains real three-b-tag (3b) collision events, 16 statistically independent replicas of synthetic four-tag (4b) events made by hemisphere mixing (each event split along the transverse thrust axis and each hemisphere replaced by its nearest kinematic neighbour from a 4b library, which destroys resonant signal but preserves detector effects), and simulated ttbar, ggF HH (kappa_lambda=1), ZH to 4b and ZZ to 4b samples. Features cover the four leading jets (pT, eta, phi, mass, DeepJet b-tag and quark-gluon scores), m4j, Higgs candidate masses, signal-region and sideband flags, trigger bits and event weights, with 191 to 623 columns depending on sample. The intended task is density-ratio estimation and domain adaptation: learn a 3b to 4b reweighting in the sideband and extrapolate to the blinded signal region, scored against mixed replica 0. About 14.5 GB, organised as data/3b, mixeddata_all/4b, ttbar/{3b,4b} and per-process folders, readable with pandas, polars or pyarrow; a quickstart notebook is included. Released by CMS under CC0-1.0.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("alefisico/LHC_HH4b_mixeddata")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `GluGluToHHTo4B_cHHH1/part_UL16_postVFP.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/8c83bc81-bef0-5d63-833a-44d99db85bea
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/8c83bc81-bef0-5d63-833a-44d99db85bea
- Over MCP: `get_schema_card` with id `8c83bc81-bef0-5d63-833a-44d99db85bea` at https://api.scigantic.com/mcp

## Related topics

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