Minimal example for HGCAL evaluation

Setup (for CPU)

# Download the model weights
xrdcp root://cmseos.fnal.gov//store/user/klijnsma/hgcal/ckpts/ckpt_train_taus_integrated_noise_Oct20_212115_best_397.pth.tar .

# Download the data and extract
xrdcp root://cmseos.fnal.gov//store/user/klijnsma/hgcal/taus_2021_v1.tar.gz .
tar xf taus_2021_v1.tar.gz

# Download the singularity container
xrdcp root://cmseos.fnal.gov//store/user/klijnsma/hgcal/pytorch_2.0.0-cuda11.7-cudnn8-devel.sif .

# Clone necessary repositories
git clone -b oc_cuda [email protected]:tklijnsma/pytorch_cmspepr.git
git clone [email protected]:tklijnsma/cmspepr_hgcal_core.git

# Boot up the container, binding current wd to /wd
singularity run --bind $PWD:/wd pytorch_2.0.0-cuda11.7-cudnn8-devel.sif

Once inside the container:

export PYTHONPATH="/opt/conda/lib/python3.10/site-packages"
cd /wd
python -m venv env
source env/bin/activate

# Install torch_geometric with extensions
pip install torch_geometric
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.0.0+cpu.html

# Install other standard packages
pip install matplotlib plotly tqdm

# Install the kNN/OC extensions for CPU
pip install -e pytorch_cmspepr/
pip install -e cmspepr_hgcal_core/

Note that pytorch is preinstalled in the container.

Usage

python plot3d.py

This script will create a file called myplots.html, which can be opened in a browser.