Download README.md from Shayla216/jet-vae-samples: direct link, hf CLI and curl.
- Browser
- Download file 3.57 kB
-
https://huggingface.co/datasets/Shayla216/jet-vae-samples/resolve/main/README.md
- Command line
-
hf download hf://datasets/Shayla216/jet-vae-samples/README.md
-
curl -L -o README.md https://huggingface.co/datasets/Shayla216/jet-vae-samples/resolve/main/README.md
license: cc-by-4.0
pretty_name: Quark and gluon jets for jet-VAE studies
tags:
- physics
- high-energy-physics
- jets
- qcd
- pythia
Quark and gluon jets for jet-VAE studies
Simulated quark and gluon jets used to train and analyse variational autoencoders of jet substructure (an image-level CNN VAE and a particle-level Transformer VAE with a Sinkhorn-EMD loss).
All jets were generated with Pythia 8.316 from a color-singlet u ubar or g g pair in its centre-of-mass frame
(invariant mass 400 GeV, 200 GeV per parton, back to back), clustered with anti-k_T (R = 0.4); the
highest-pT jet is kept. Particle-identification information is not used by the models.
Contents
| Path | What | Size |
|---|---|---|
training_jets/u_jet.dat, training_jets/g_jet.dat |
100 000 quark jets and 100 000 gluon jets at the default Pythia settings; the samples used to train the particle-level VAE and as reference jets | 120 MB, 180 MB |
pythia_scan/200GeV_process/ |
the parameter scan: 10 000 quark and 10 000 gluon jets for each of 33 settings | 642 MB |
SHA256SUMS.txt |
checksums of every data file |
training_jets/*.dat
Plain text, one jet per block:
<event_begin>:id
pid px py pz E
...
<event_end>:id
One row per constituent (PDG id, momentum components and energy in GeV). The order of the jets in the file is the event index (for example event 4 of the quark file).
pythia_scan/200GeV_process/
{u_jet,g_jet}/nominal.npz default settings
{u_jet,g_jet}/<parameter>/{minus_20pct,minus_10pct,plus_10pct,plus_20pct}.npz
Eight parameters, each varied by -20%, -10%, +10%, +20% relative to its default (one at a time):
| Folder | Pythia setting | Default |
|---|---|---|
alphaS |
TimeShower:alphaSvalue |
0.1365 |
aLund |
StringZ:aLund |
0.68 |
bLund |
StringZ:bLund |
0.98 |
sigma |
StringPT:sigma |
0.335 |
pTmin |
TimeShower:pTmin |
0.5 |
mesonUDvector |
StringFlav:mesonUDvector |
0.5 |
probStoUD |
StringFlav:probStoUD |
0.217 |
probQQtoQ |
StringFlav:probQQtoQ |
0.081 |
Each .npz holds 10 000 jets (u_jet = quark, g_jet = gluon) with these arrays:
| Array | Shape | Meaning |
|---|---|---|
particles80 |
(10000, 80, 3) | (pT fraction, delta_eta, delta_phi) of the 80 leading constituents, ordered by pT, centred on the jet axis; the input of the particle-level VAE |
mask80 |
(10000, 80) | True for real constituents (the rest is padding) |
particles, mask |
(10000, K, 3), (10000, K) | the same without truncation (K = longest jet in the file) |
p4 |
(10000, K, 4) | px, py, pz, E in GeV, before centring; used to build the images of the CNN VAE |
pid |
(10000, K) | PDG ids aligned with p4 |
y |
(10000,) | 0 = gluon, 1 = quark |
n_constituents, scalar_pt_GeV, original_jet_p4 |
per-jet summaries of the selected jet | |
source_index, seed, event_id, shard_index |
bookkeeping | |
metadata |
scalar string | JSON with dataset_id, parameter, relative_change, the Pythia settings, the preprocessing description and label definition |
Different scan points are not event-paired. p4 and pid are the analysis-input constituents, not unmodified
generator output.
import json, numpy as np
d = np.load("pythia_scan/200GeV_process/g_jet/alphaS/plus_10pct.npz")
meta = json.loads(str(d["metadata"])) # meta["parameter"], meta["relative_change"], ...
jets, mask = d["particles80"], d["mask80"]
License
CC-BY-4.0.