This is a Chainer implementation of Defferrard et al., "Convolutional Neural Networks on Graphs with Fast Localized Spectral Filtering", NIPS 2016.
Use it at your own risk. See license for details.
This is not the original author's implementation. This implementation was based on https://github.com/mdeff/cnn_graph.
I updated original repository to work on chainer v5.
# Trains a GraphCNN on MNIST
$ python tools/train.py -c configs/default.json -o results -e 100 -g 0
python 3.5.5
chainer (5.1.0)
cupy (5.1.0)
nose (1.3.7)
scikit-learn (0.19.2)
scipy (1.1.0)
MIT License. Please see the LICENSE file for details.