bwittmann/gav

A link prediction algorithm tailored to flow-driven spatial networks. Paper accepted @ WACV24

Graph Attentive Vectors (GAV)

To run our proposed GAV framework on the ogbl-vessel benchmark, please follow the instructions below. The ogbl-vessel benchmark's data will be automatically downloaded and stored under ./dataset. For details, please check our paper and its extensive supplementary material.

News

January 24: Link Prediction for Flow-Driven Spatial Networks has been accepted and presented at WACV!

Installation

Please create a new virtual environment using, e.g., anaconda:

conda create --name gav python=3.8.15

Subsequently, install the required packages:

pip install torch==1.13.1 --extra-index-url https://download.pytorch.org/whl/cu116

pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv torch_geometric -f https://data.pyg.org/whl/torch-1.13.0+cu116.html

pip install ogb tensorboard tqdm networkit

The installation was tested using Ubuntu 16.04 and CUDA 11.6.

Training

To train GAV, please run:

python gav_link_pred.py --save_appendix <appendix> --gpu_id <gpu_id> --dataset ogbl-vessel

Checkpoints and tensorboard log files will be stored under ./results.

Testing

To test GAV's performance on an individual checkpoint, please run:

python gav_link_pred.py --save_appendix <appendix_of_run_of_ckpt> --gpu_id <gpu_id> --dataset ogbl-vessel --only_test --continue_from <ckpt_epoch_nr>

To test GAV's performance on our provided checkpoint, please run:

python gav_link_pred.py --save_appendix _gav --gpu_id <gpu_id> --dataset ogbl-vessel --only_test --continue_from 34

Road Networks

Preprocessing

To preprocess the road network datasets, please download the graph and coordinates from here. The downloaded file from Graph should be called edges.graph, while the downloaded file from Coordinates should be called nodes.graph. Both files should be stored in an individual directory located at <path_to_downloaded_files>. Finally, run:

python create_dataset.py --path <path_to_downloaded_files> --gpu_id <gpu_id> --dataset_name <e.g., ogbl-luxembourg_road>

Please note that in the preprocessing step, the --dataset_name has to start with ogbl- and should not include additional hyphens.

Training and Testing

Follow the instructions above and simply state the processed dataset's name after the --dataset flag, omitting ogbl-. E.g., --dataset luxembourg_road.

More Whole-Brain Vessel Graphs

Preprocessing

To preprocess additional whole-brain vessel graphs, please download the raw data from here. The downloaded files should be stored in an individual directory located at <path_to_downloaded_files> Finally, run:

python create_dataset.py --path <path_to_downloaded_files> --gpu_id <gpu_id> --dataset_name <e.g., ogbl-c57_tc_vessel>

Please note that in the preprocessing step, the --dataset_name has to start with ogbl- and should not include additional hyphens.

Training and Testing

Follow the instructions above and simply state the processed dataset's name after the --dataset flag, omitting ogbl-. E.g., --dataset c57_tc_vessel.

Citation

If you find our work useful for your research, please consider citing:

@InProceedings{Wittmann_2024_WACV,
    author    = {Wittmann, Bastian and Paetzold, Johannes C. and Prabhakar, Chinmay and Rueckert, Daniel and Menze, Bjoern},
    title     = {Link Prediction for Flow-Driven Spatial Networks},
    booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
    month     = {January},
    year      = {2024},
    pages     = {2472-2481}
}