Our code was tested on Python 3.11 on windows and Ubuntu (Cuda 12.6)
using pip with the “requirements-all.txt” should be enough.
If you want the exact versions of the packages, uses “requirements-exact.txt” found here: https://figshare.com/articles/software/Pip_requirements_with_exact_versions/29135252
The dataset is available on Hugging Face on Hugging Face
LISTTT/NeurIPS_2025_BMDB at main
The pre-trained models are available on https://figshare.com/projects/SafePowerGraph/248768
They are organized in folders for each case size then each graph architecture; the name of the files follow the format: {start_lr}{case_size}{architecture}{training_case}{opf}{pinn}{validation_case}_{weighting}.pt example: 0.01_case9_sage_load_relative_1_2_2_load_relative_uniform.pt
To enable experiment tracking with comet (www.comet.com) set the environment variable COMET_APIKEY
export COMET_APIKEY=XXX
To enable Hugging Face dataset loading, set the environment variable HUGGINFACE_TOKEN
export HUGGINFACE_TOKEN=XXX