PocketCoffea VHcc setup

1. Install Miniconda or Micromamba
   • At RWTH it should be at your /net/scratch_cms3a/<username> area
   • At lxplus CERN, it should be in your /eos/user/u/username area
2. Create a dedicated environment for PocketCoffea, install the packages, and compile:

   conda create -n PocketCoffea python=3.10 -c conda-forge
   conda activate PocketCoffea
   # install PocketCoffea
   git clone [email protected]:PocketCoffea/PocketCoffea.git
   cd PocketCoffea
   pip install -e .
   

   Follow their installation instructions for other options. Afterwards install additional packages needed for BDT/DNN training and evaluation. Keep using conda, since using pip might alter the environment, leading to conflicts.

   conda install conda-forge::xrootd
   conda install conda-forge::lightgbm
   conda install conda-forge::tensorflow
   conda install pytorch::pytorch
   conda install setuptools==70.*    
   

   For brux20 cluster at Brown, you may need conda install conda-forge::ca-certificates.
3. Checkout this repo:

   git clone [email protected]:cms-rwth/VHccPoCo.git
   

4. (If your local username is different from your CERN username) Setup your CERN username variable:

   export CERN_USERNAME="YOURUSERNAME"
   

5. Follow examples to create dataset input files. First activate voms proxy. Then:

   cd VHccPoCo
   mkdir datasets
   build-datasets --cfg samples_Run3.json -o -ws T2_DE_RWTH -ws T2_DE_DESY -ws T1_DE_KIT_Disk -ws T2_CH_CERN -ir
   cd ../
   

   Use -p 12 with build-datasets to parallelizing with 12 cores, e.g.
6. Run with the futures executor (test before large submission):

   runner --cfg VHccPoCo/cfg_VHcc_ZLL.py -o output_VHcc_Test --executor futures -s 10 -lf 1 -lc 1
   

7. Run on condor with Parsl executor (only if the previous step was successeful):

   runner --cfg VHccPoCo/cfg_VHcc_ZLL.py -o output_VHcc_v01 --executor parsl-condor@RWTH -s 60
   

8. Make some plots:

   make-plots -inp output_VHcc_v01 -op VHccPoCo/params/plotting.yaml
   

   The plot parameters can be changed by editing VHccPoCo/params/plotting.yaml.