Cluster-GCN in PyTorch

Cluster-GCN: An Efficient Algorithm for Training Deep and Large Graph Convolutional Networks Wei-Lin Chiang, Xuanqing Liu, Si Si, Yang Li, Samy Bengio, Cho-Jui Hsieh. KDD, 2019. [Paper] Raw data files used to curate this dataset can be downloaded from http://manikvarma.org/downloads/XC/XMLRepository.html while the processed data files used in this implementation can be downloaded by clicking on the above Download amazon2M badge.

GraphConv Layer module usage

from layers import GraphConv
gconv = GraphConv(in_features: int, out_features: int)
out = gconv(A: torch.sparse.Floattensor, X: torch.FloatTensor) # out.shape(N, out_features)

where, N = number of nodes in the graph
F = number of features per node
in_features = number of input features in X,
out_features = number of output features,
X = a dense FloatTensor containing input features,
A = a sparse FloatTensor representing the graph adjacency matrix created as follows

i = torch.LongTensor(indices)
v = torch.FloatTensor(values)
A = torch.sparse.FloatTensor(i.t(), v, shape)

Requirements:

• install the clustering toolkit metis and other required Python packages.

1) Download metis-5.1.0.tar.gz from http://glaros.dtc.umn.edu/gkhome/metis/metis/download and unpack it
2) cd metis-5.1.0
3) make config shared=1 prefix=~/.local/
4) make install
5) export METIS_DLL=~/.local/lib/libmetis.so
6) pip install -r requirements.txt

Usage:

• Run the below shell script to perform experiments on amazon2M dataset.

./amazon2M.sh
tensorboard --logdir=runs --bind_all (optional to visualize training)

NOTE: The Scipt assumes that the data files are stored in the following structure. 
./datasets/amazon2M/amazon2M-{G.json, feats.npy, class_map.json, id_map.json}

Results:

• Test F1 0.8880 (vs Cluster-GCN paper - 0.9041)
• Training Loss 0.3096

• Training Accuracy 0.9021