Continuous-Time Graph Anti-Symmetric Network (CTAN).

Official reference implementation of our paper "Long Range Propagation on Continuous-Time Dynamic Graphs" accepted at ICML 2024, and "Effective Non-Dissipative Propagation for Continuous-Time Dynamic Graphs" accepted at the Temporal Graph Learning Workshop @ NeurIPS 2023

Please consider citing us

  @InProceedings{gravina2024ctan,
    title = {Long Range Propagation on Continuous-Time Dynamic Graphs},
    author = {Gravina, Alessio and Lovisotto, Giulio and Gallicchio, Claudio and Bacciu, Davide and Grohnfeldt, Claas},
    booktitle = {Proceedings of the 41st International Conference on Machine Learning},
    pages = {16206--16225},
    year = {2024},
    editor = {Salakhutdinov, Ruslan and Kolter, Zico and Heller, Katherine and Weller, Adrian and Oliver, Nuria and Scarlett, Jonathan and Berkenkamp, Felix},
    volume = {235},
    series = {Proceedings of Machine Learning Research},
    month = {21--27 Jul},
    publisher = {PMLR},
    pdf = {https://raw.githubusercontent.com/mlresearch/v235/main/assets/gravina24a/gravina24a.pdf},
    url = {https://proceedings.mlr.press/v235/gravina24a.html},
  }

Credits

Many thanks to Giulio Lovisotto (Github / Homepage) for his invaluable help on this project.

Requirements

Note: we assume Miniconda/Anaconda is installed, otherwise see this link for correct installation. The proper Python version is installed during the first step of the following procedure.

There are two ways to install the required packages:

1. Run the script create_env.sh

or

2. Run conda env create -f env.yml followed by conda activate ctan

Then

cd TGB; python3 -m pip install .

How to reproduce our results

• Sequence Classification on Temporal Path Graph

  python3 run_sequence.py

• Classification on Temporal Pascal-VOC

  python3 run_pascal.py

• Future link prediction tasks (ie, Wikipedia, Reddit, LsatFM, MOOC)

  python3 run_link_prediction.py

• Temporal Graph Benchmark

  python3 run_tgb.py

Usage

This repo builds on CTDG-learning-framework, a framework to easlity experiment with Graph Neural Networks (GNNs) in the temporal domain. Please refer to it for more details.

Repository structure

The repository is structured as follows:

├── README.md                <- The top-level README.
│
├── env.yml                  <- The conda environment requirements.
├── create_env.sh            <- The script for building the environment.
│
├── main.py                  <- The main file for Sequence Classification on Temporal Path Graph, Classification on Temporal Pascal-VOC, and Future link prediction tasks.
├── main_tgb_ctan.py         <- The main file for the Temporal Graph Benchmark.
├── negative sampler.py      <- Implements the class of the negative sampler.
├── train_link.py            <- Implements the code responsible for training and evaluation of the models in link prediction tasks.
├── train_sequence.py        <- Implements the code responsible for training and evaluation of the models in sequence prediction tasks.
├── utils.py                 <- Contains the code for multiple utilities.
│
├── run_link_prediction.py   <- The script for reproducing the Future link prediction results.
├── run_sequence.py          <- The script for reproducing the Sequence Classification on Temporal Path Graph.
├── run_pascal.py            <- The script for reproducing the Temporal Pascal-VOC results.
├── run_tgb.py               <- The script for reproducing the Temporal Graph Benchmark results.
│
├── confs                    <- Contains the the evaluated model configurations.
│      
├── data                     <- Contains the data for Sequence Classification on Temporal Path Graph and Classification on Temporal Pascal-VOC tasks.
│      
├── datasets                 <- Contains the code to create and load the datasets.
│      
├── models                   <- Contains the code to implement CTAN and other C-TDG GNNs.
│      
└── TGB                      <- Contains the code of the TGB utilities.