deep-mining-swang/role-action-embeddings

Code for paper on role action embeddings

Please cite as

Berry, George. "Role action embeddings: scalable representation of network positions." arXiv preprint arXiv:1811.08019 (2018).

Notes

This was created using Python 3, it may not work with Python 2. All packages were installed using Anaconda's miniconda distribution.

We include slightly modified versions of the node2vec and GraphWave repos for comparison.

We use model loading code from Kipf and Welling.

Graph classification datasets were downloaded here.

Node classification datasets can be found here, but we use the specific splits provided here.

Model code was adapted from Hamilton et al's code.

Make sure you have the dependencies installed. These are: scikit-learn, numpy, pandas, networkx, matplotlib, seaborn, pytorch, gensim, jupyter, scipy

How to run small_graphs.ipynb code

• Open a Jupyter notebook with jupyter notebook at the command line and open small_graphs.ipynb

How to run code in the empirical_graphs folder

• There are four notebooks, one for Cora, one for Citeseer, one for Pubmed, and one for all of the graph classification tasks
• Unzip empirical_graphs/data.zip
• empirical_graphs/models.py, you should set the base_data_path variable to point to the data folder on your machine. Data is contained in the empirical_graphs/data folder
• For the graph classification, the REDDIT-MULTI-12k files are quite large, please download it from here.

What are these old- files doing?

• They're leftovers from my original runs to generate the results in the paper. I then went back and cleaned up the code. You can ignore them, but I wanted to keep them around for posterity.