1. Source softwarestack from LCG (always needed)

    source /cvmfs/sft.cern.ch/lcg/views/LCG_93c/x86_64-slc6-gcc62-opt/setup.sh 
   

2. Install xgboost and root_pands (only needed once)

    pip install --user xgboost 
    pip install --user keras==2.1.1
    pip install --user root_pandas==0.6.1 
   

3. Add local python packages to PYTHONPATH ( best to add to ~/.profile )

    `export PYTHONPATH=$HOME/.local/lib/python2.7/site-packages:$PYTHONPATH`
   

Set up a CMSSW environment (>=9_4_0). All needed packages are included in the softwarestack.

However, keras and root_pandas need to be updated to version 2.1.1 and 0.6.1 respectively

Use run_model.py to train (-t) model (-m [keras,xgb]) on events for a given channel (-c [mt,et,tt]). Samples and input variables must be specified in conf/global_config*.json. To get proper training weights run calc_train_weights.py after variables and samples are specified.