Official code repository for PAKDD 2023 paper "Fast and Attributed Change Detection on Dynamic Graphs with Density of States"
We include historydosN20.pkl which is the computed DOS embedding and can be used with Anomaly_Detection.py
similarly, we include skynet_gdos.pkl and china.pkl to reproduce COVID flight network experiment
run datasets/multi_SBM/SBM_generator.py to generate SBM hybrid experiments
run datasets/multi_SBM/SBM_addnode.py to generate SBM Evolving Size experiment
in subroutines/ADOS/run_ADOS.mat to run attributed DOS with LDOS
follow main function in dos.py to generate dos embeddings in python, and for real world experiments
follow main function in spotlight.py to run SPOTLIGHT experiments
To use local DOS to approximate eigenvectors
You can use the MATLAB code under SCPD/subroutines/ADOS.
Because the interaction with eigenvectors are approximated with the GQL approximation method (currently only implemented in MATLAB so far).
Feel free to reach out to me if you have any questions: shenyang.huang@mail.mcgill.ca
If code or data from this repo is useful for your project, please consider citing our paper:
@inproceedings{huang2023fast,
title={Fast and Attributed Change Detection on Dynamic Graphs with Density of States},
author={Huang, Shenyang and Danovitch, Jacob and Rabusseau, Guillaume and Rabbany, Reihaneh},
booktitle={Pacific-Asia Conference on Knowledge Discovery and Data Mining},
pages={15--26},
year={2023},
organization={Springer}
}