DeepMETv1 is a fully-connected neural network (FCNN) for MET reconstruction in CMS data. This branch is focused on training on Run3 conditions.
The original repository by Yongbing Feng was used to train DeepMETv1 on Run2 conditions. A copy of this original code is in the branch Run2.
Install the necessary packages with MiniConda. You can use the provided env_deepmetv1.yml file.
conda env create -y -f env_deepmetv1.yml
Activate the environment
conda activate METTraining
Prepare the HDF5 training files (to see log info, add option -v)
python convertNanoToHDF5.py inputfile.root outputfile.h5
To see more options
python convertNanoToHDF5.py --help
Run the training
python train_ptmiss_mine.py -i input.txt