In the following example, we take a Delphes ROOT file events_delphes_example.root as input, run an analyzer macro analyze.C to infer the Sophon model, and output both scores and latent space features to a new ROOT file out.root.
This example works on EL9 machines.
# Setup environment
source /cvmfs/sft.cern.ch/lcg/views/LCG_104/x86_64-el9-gcc13-opt/setup.sh
export ROOT_INCLUDE_PATH=$ROOT_INCLUDE_PATH:/cvmfs/sft.cern.ch/lcg/releases/delphes/3.5.1pre09-9fe9c/x86_64-el9-gcc13-opt/include:/cvmfs/sft.cern.ch/lcg/releases/onnxruntime/1.15.1-8b3a0/x86_64-el9-gcc13-opt/include/core/session
# Download Sophon model
wget --no-verbose https://huggingface.co/jet-universe/sophon/resolve/main/models/JetClassII_Sophon/model.onnx -O JetClassII_Sophon.onnx
# Compile and run macro
root -b -q 'analyze.C++("events_delphes_example.root", "out.root", "JetClassII_Sophon.onnx")'Note:
To ensure that the Sophon model achieves the expected performance, it is highly recommended that the Delphes file is produced from the delphes_card_CMS_JetClassII card series provided in the jetclass2_generation repository.
This card is based on the CMS detector configuration, applying an additional track smearing, emulating the PU effect with