Heuristics for Dynamic Graph Link Prediction

The main content of this project lies in the simple_run_scripts.

Run

To run the scripts, you can run_experiments.py for the TGB workloads and run_benchtemp_exp_with_sampled_metrics.py for BenchTemp.

Create your own (stacked) heuristic!

If you wish to create your own heuristic, make sure that it:

• Implements all the functions of the abstract base class Tracker in trackers/base.py.
• Gives a score which is an integer (a natural number) with a max value (assuming you wish to utilize the FenwickTree ranking technique).

To create your own stacked heuristic, you can do:

from simple_run_scripts.trackers.global_tracker import GlobalRecencyTracker
from simple_run_scripts.trackers.popularity_tracker import GlobalPopularityTracker
from simple_run_scripts.trackers.combination_ranker import CombinationTracker

# Create [GR, GP] ranker
class CustomTracker(CombinationTracker):
    def __init__(self, all_timestamps, num_nodes, num_edges):
        trackers = [
            GlobalRecencyTracker(all_timestamps, num_nodes, num_edges),
            GlobalPopularityTracker(all_timestamps, num_nodes, num_edges),
        ]
        super().__init__(
            all_timestamps, 
            num_nodes, 
            num_edges, 
            trackers_list=trackers
        )

Installation

Please make sure that you have installed:

• TGB (pip install py-tgb)
• Benchtemp (pip install benchtemp)
• torch_geometric installed (see installation instructions here)

Citing

Upon using this repository for your work, or finding our proposed analysis useful for your research, please consider citing our paper this paper:

@InProceedings{pmlr-v296-cornell25a,
  title = {On the Power of Heuristics in Temporal Graphs},
  author = {Cornell, Filip and Smirnov, Oleg and Zarzar Gandler, Gabriela and Cao, Lele},
  booktitle = {Proceedings on "I Can't Believe It's Not Better: Challenges in Applied Deep Learning" at ICLR 2025 Workshops},
  pages = {37--46},
  year = {2025},
  volume = {296},
  series = {Proceedings of Machine Learning Research},
  month = {28 Apr},
  publisher = {PMLR},
  url = {https://proceedings.mlr.press/v296/cornell25a.html},
}