Manipulate collision events via graphs and graph neural networks.
The EventsDataset, a Pytorch Geometric Dataset, allows you to download a dataset of graphs representing collisions.
from dataset import EventsDataset
graphs = EventsDataset(
root='/Users/alessiodevoto/Desktop/test_dataset3',
delete_raw_archive=False,
event_subsets={'signal': 100, 'singletop': 100, 'ttbar': 100})
graphsEventsDataset(300)
Each event is a graph with 6/7 nodes. Each node is built from the raw file as follows:
| Particle | Feature 1 | Feature 2 | Feature 3 | Feature 4 | Feature 5 | Feature 6 |
|---|---|---|---|---|---|---|
| jet1 | 'pTj1' | 'etaj1' | 'phij1' | 'j1_quantile' | nan | nan |
| jet2 | 'pTj2' | 'etaj2' | 'phij2' | 'j2_quantile' | nan | nan |
| jet3 (optional) | 'pTj3' | 'etaj3' | 'phij3' | 'j3_quantile' | nan | nan |
| b1 | 'pTb1' | 'etab1' | 'phib1' | 'b1_quantile' | 'b1m' | nan |
| b2 | 'pTb2' | 'etab2' | 'phib2' | 'b2_quantile' | 'b2m' | nan |
| lepton | 'pTl1' | 'etal1' | 'phil1' | nan | nan | nan |
| energy | 'ETMiss' | nan | 'ETMissPhi' | nan | nan | 'metsig_New' |
g = graphs[0]
print(g)Data(x=[6, 6], edge_index=[2, 30], y=[1], event_id='signal_6350')
from visualize import plot_event_2d
plot_event_2d(graphs[100])
a_list_of_graphs = [graphs[i] for i in range(0, 300, 30)]
plot_event_2d(a_list_of_graphs, height=1500)
@article{genovese2025mixtureofexpertsgraphtransformersinterpretable,
title={Mixture-of-Experts Graph Transformers for Interpretable Particle Collision Detection},
author={Donatella Genovese and Alessandro Sgroi and Alessio Devoto and Samuel Valentine and Lennox Wood and Cristiano Sebastiani and Stefano Giagu and Monica D'Onofrio and Simone Scardapane},
year={2025},
url={https://arxiv.org/abs/2501.03432},
}
- version 0.0.5. New dataset.
- version 0.0.4.3. Now the plotting function allows you to pass a show_edges and edges_weights parameters to display edges. Useful for visualizing attention maps.