/l1-jet-id

Training, testing and synthetisation for FPGA of three machine learning models: MLP, deepsets, and interaction networks with application to HL-LHC jet tagging.

Primary LanguagePythonApache License 2.0Apache-2.0

l1-jet-id

DOI

Code for Level-1 jet tagging

Data

Raw data: https://zenodo.org/record/3602260#.YnT0xZpBz0o

Kfolded data (k=10) with each consituent having transverse momentum larger than 2 GeV (pT > 2 GeV): https://cernbox.cern.ch/s/L714Jx0ENCg8uTd

The 5 jet classes have equal representation in each kfold. The jets in every kfold are shuffled using different seeds. Moreover, the constituents of each jet are shuffled using a different seed.

For robust normalisation data, the parameters are:

x_median: [13.901342868804932, 0.0, 0.0]
interquantile_range: [119.6879810333252, 0.27309816256165514, 0.27362077757716186]

K-Folding

Consider the 5 folds (k=5) and label them: 1, 2, 3, 4, 5

[1, 2, 3, 4] for training + [5] for validation <- use this one for hyperparameter optimization (HO)

[2, 3, 4, 5] for training + [1] for validation

[3, 4, 5, 1] for training + [2] for validation

[4, 5, 1, 2] for training + [3] for validation

[5, 1, 2, 3] for training + [4] for validation