Performance of model training for ttH-Multilepton analysis (2lSS)

Prepare samples:

Skimm root ntuples

Follow instructions from skim: Output is lightweight root files with weights and input variables, could be found at Files/skimmed. ttH - signal, ttW - background.

Convert root to flat file

Compare various options to access root files: format_change:

• access_root - use uproot and pandas
• ?!

Model training :

model_training - contains various sets of training for ttH vs ttW.

• standard_tmva - nominal training using C++:
  • Basic functional - read trees, train (not include application)
• tmva_python - python based use of TMVA, tthml_TMVAtraining_python.ipynb:
  • Basic functional
    • Reproduce "nominal"
  • Introduce new features:
    • Cross Validation
    • Plot combined ROC curves
    • Test NN implementation
• ml_packages - Use set of various industry-conventional tools:
  • Training setups
    • Weak learners (different set of BDTs)
    • Neural Networks
    • Introduce multiclass for ttH
  • Access setups
    • dataframes, spark