Awesome

Gae In Pytorch

Graph Auto-Encoder in PyTorch

This is a PyTorch/Pyro implementation of the Variational Graph Auto-Encoder model described in the paper:

T. N. Kipf, M. Welling, Variational Graph Auto-Encoders, NIPS Workshop on Bayesian Deep Learning (2016)

This repository uses some of the code found here: https://github.com/tkipf/pygcn and https://github.com/tkipf/gae.

Tested December 19th, 2018 with PyTorch 1.0 and Pyro 0.3.0.

Requirements

• Python 2.7
• Pyro 0.3.0
• PyTorch 1.0
• networkx
• scikit-learn
• scipy
• numpy
• matplotlib
• pickle

To run

After installing all requirements:

python train.py

Notes

• This implementation uses Pyro's blackbox SVI function with the default ELBO loss. This is slower than the TensorFlow implementation which uses a custom loss function with an analytic solution to the KL divergence term.
• Currently the code is not set up to use a GPU, but the code should be easy to extend to improve running speed