To do

Below is a list of items which should ideally be addressed as part of the development and study of de-correlated tagger(s). Developers should feel free to assign themselves to any task they see fit.

PUBNOTE

• High priority
  • Ntuple production
    • Rel. 21 MC ntuples submitted
    • Rel. 21 MC ntuples ready
  • Adversarial training
    • Determine convergence condition for adversarial training
  • Adaboost/uBoost optimisation (@gmarceca)
  • Final classifier optimisation (awaiting Rel. 21 ntuples)
• Low priority
  • Study adversarial training with flatness re-weighting
  • Determine scheme for feature selection
  • Parametrised training
    • Request varied-mass signal samples in JETM6 (@gmarceca)
      • Prodtask link: https://prodtask-dev.cern.ch/prodtask/inputlist_with_request/15428/
    • Run ntuple-dumper on varied-mass signal samples (@gmarceca)
    • Implement parametrised training (@gmarceca)

Data studies

• Implementation in BoostedJetTaggers
• Get data/MC files from W/top-tagging CONF/paper
• Test on e.g. W/Z-peak in data