hookk / graph-nn

Graph Convolutional Neural Networks for Theano

Graph Convolutional Nets

Graph Convolutional Neural Networks for Theano based on (kpif et al.) for classification on the baseline datasets considered in their paper. These datasets are included in the repository in the data folder.

Dependencies

• theano
• networkx (to unpack the dataset)

Running the code(demo)

To train the model, simple do

python -W ignore main.py

This will save a model after every 500 iterations and also test the model on the test dataset. A saved model can be loaded using the --lm option as

python -W ignore main.py --lm=$NameSavedModel

Finally to test a saved model it can be done with the --t flag as.

python -W ignore main.py --lm=$NameSavedModel --t=0

The model achieves good accuracies on baseline datasets

Dataset	Accuracy
Cora	80.5
Citeseer	71.1
Pubmed	79.2
NELL	65.8