Interested people: Loukas, Huilin, Anna, Roman ==== Ideas ==== - train W-vs-q/g and q-vs-g with state-of-the-art generators - consider training dijet vs Z+jet vs W(qq) without any use of "flavor-truth" to remain generator independent. cross check with CMS-style-parton-flavor-definition - reweight to different observables in measurements, check performance - make a rivet plugin of a ML-tagger as testbed to MC generators - Identify which part of the hadronization process shows the largest discrepancy between the generators - Cross-correlation between the discriminate of the same algorithm training using different MC generators. e.g. jet tagged with one MC generator, not tagged by the other - Change some parameters of pythia's hadronization model. Can a classifier discriminate between those and help tuning More in: More in https://www.overleaf.com/project/648ab3e1c164ede47c68c368 ==== Rivet plugin to dump a root tree ==== # setup LCG environment source /cvmfs/sft.cern.ch/lcg/views/LCG_103/x86_64-centos7-gcc11-opt/setup.sh # build example analysis wget https://gitlab.com/hepcedar/rivet/-/raw/rivet-3.1.7/analyses/examples/EXAMPLE_NTUPLE_ROOT.cc rivet-build RivetExampleAnalyses.so EXAMPLE_NTUPLE_ROOT.cc `root-config --cflags --libs` export RIVET_ANALYSIS_PATH=$PWD # check the build is successful rivet --show-analysis EXAMPLE_NTUPLE_ROOT # run it rivet --analysis EXAMPLE_NTUPLE_ROOT 0000.hepmc Start from example https://gitlab.com/hepcedar/rivet/-/blob/release-3-1-x/analyses/examples/EXAMPLE_NTUPLE_ROOT.cc First version from Huilin: https://github.com/lh23-jss/lh23-jss-rivet JET_NTUPLE_QG:MODE=DIJET:JET_R=0.4 for q-vs-g JET_NTUPLE_QG:MODE=ZJET:JET_R=0.4 for q-vs-g JET_NTUPLE_QG:MODE=DIJET:JET_R=0.8 for W-vs-q/g JET_NTUPLE_QG:MODE=WZ:JET_R=0.8 for W-vs-q/g ==== MC samples ==== At hadron level with all particles HEPMC Do not need the whole event, but only the particles within the leading jet List of samples and generators, see https://phystev.cnrs.fr/wiki/2023:groups:smjets:jss-measurements:start