danielegrattarola / cdt-ccm-aae

Change Detection in Graph Streams by Learning Graph Embeddings on Constant-Curvature Manifolds (2018, https://arxiv.org/abs/1805.06299)

This is the official implementation of the paper "Change Detection in Graph Streams by Learning Graph Embeddings on Constant-Curvature Manifolds" by D. Grattarola, D. Zambon, C. Alippi, and L. Livi. (2018, https://arxiv.org/abs/1805.06299).

This code provides a proof of concept for the experiments conducted in the paper, and contains all the necessary elements to apply the methodology on a generic problem.

Please cite the paper if you use any of this code for your research:

@article{grattarola2019change,
  title={Change detection in graph streams by learning graph embeddings on constant-curvature manifolds},
  author={Grattarola, Daniele and Zambon, Daniele and Livi, Lorenzo and Alippi, Cesare},
  journal={IEEE transactions on neural networks and learning systems},
  year={2019},
  publisher={IEEE}
}

Setting up

The code is implemented for Python 3 and tested on Ubuntu 16.04.
To run the code, you will need a number of libraries installed on your system:

• Keras (pip install keras), a high-level API for deep learning;
• Spektral (pip install spektral), a Keras extension to build graph neural networks;
• CDG (see README on Github), a library for change detection tests and non-Euclidean geometry.

The code also depends on Numpy, Scipy, Pandas, Scikit-learn, and Joblib.

Running experiments

The src folder includes several scripts to run the different versions of the algorithm proposed in the paper:

• main_prior.py, trains the AAE using the non-Euclidean prior;
• main_geom.py, trains the AAE using the geometric discriminator;
• main_cdt.py, runs the change detection tests on the data saved by the previous scripts;

To simplify the workflow, the repository includes a main.sh script to run a full experiment on Delaunay triangulations.

About

For a simplified explanation of the paper, check out this blog post.