License:

MIT License

Documentation

Extensive documentation is provided in https://ballistic-purple-6f5.notion.site/SafePowerGraph-Documentation-1fcbf710ca5980fdafccccf2a9891fc3

Quick Start Examples

The file "example_script.sh" provides a list of different commands with examples of the parameters;

Third party tools:

Trackers:

To enable experiment tracking with comet (www.comet.com) set the environment variable COMET_APIKEY

export COMET_APIKEY=XXX

Models:

We provide A model zoo of pretrained models using different weighting, architecture, and pinn strategies They are organized by grid size and available on: https://figshare.com/projects/SafePowerGraph/248768

If you just want to evaluate an existing model and not fine-tune it, set the argument --epochs to 0 and provide in --model_file the .pt file of the weights you downloaded

Datasets:

• To enable Hugging Face to load OUR datasets, set the environment variable HUGGINFACE_TOKEN

export HUGGINFACE_TOKEN=XXX

The files in Hugging face are available on: https://huggingface.co/datasets/LISTTT/NeurIPS_2025_BMDB/tree/main Ensure that you are using the same seed (the xxx just before .pkl in the link) so it can be loaded properly. We hosted in hugging face the seeds 20, 25, and 200.

Our datasets are Licensed under the Creative Commons Attribution 4.0

• We support the OPF Dataset in our benchmark https://pytorch-geometric.readthedocs.io/en/latest/generated/torch_geometric.datasets.OPFDataset.html#torch_geometric.datasets.OPFDataset with grids up to 13659 bus to use them, train and evaluate the models using the script in experiments/train_eval_largeOPF.py

This dataset is also licensed under the Creative Commons Attribution 4.0

For more information on this dataset, refer to their README: https://storage.mtls.cloud.google.com/gridopt-dataset/README