Graph convolutional networks

Overview

My attempt to reproduce graph classification results from recent papers [1, 2] using Graph U-Net. So far, my results using Graph U-Net are worse than the baseline (GCN).

I also compare to our recent work on Multigraph GCN (MGCN) and Multigraph ChebNet [4]. More results are presented in Table 1 of [4].

Citation

Please use this citation if you want to cite this code in your paper:

@misc{knyazev2018graphnn,
  author = {Knyazev, Boris},
  title = {Graph Classification with Graph Convolutional Networks in PyTorch},
  year = {2018}
}

Update:

• Added support of one-hot node degree features by using the --degree flag.
• Added support of the COLORS and TRIANGLES datasets from our recent paper. Note that many features like unsupervised/supervised attention are not supported in this repo, see the original graph_attention_pool repo for all experiments.

Examples:

python graph_unet.py -D COLORS-3 --folds 1 -gc
python graph_unet.py -D TRIANGLES --folds 1 -gc --degree

Data

This repository contains all necessary data for the PROTEINS dataset. It can be found here along with similar datasets or in the data folder of this repo.

Dataset statistics and references to papers are available at here.

Models

The baseline model is Graph Convolutional Network (GCN) [3]. The decoder part of Graph U-Net is not implemented yet in our code, i.e. the only difference with the baseline is using pooling based on dropping nodes between graph convolution layers.

Hyperparameters are taken from [2], but learning rate decay and dropout is also applied. The readout layer (last pooling layer over nodes) is also simplified to just max pooling over nodes. All hyperparameters are the same for the baseline GCN, Graph U-Net and Multigraph GCN (MGCN) except for the last row in the tables, in which case hyperparameters from [4] are used.

Implementation is very basic without much optimization, so that it is easier to debug and play around with the code.

python graph_unet.py --model gcn  # to run baseline GCN
python graph_unet.py --model unet  # to run Graph U-Net
python graph_unet.py --model mgcn  # to run Multigraph GCN
python graph_unet.py --model mgcn -K 2  # to run Multigraph ChebNet with filter scale K = 2

To use the PyTorch Geometric data loader, add flag --torch-geom or -g.

Repeating 10 times for different seeds (as we do in our paper [4]):

for i in $(seq 1 10); do seed=$(( ( RANDOM % 10000 )  + 1 )); python graph_unet.py --model gcn --seed $seed | tee logs/gcn_proteins_"$i".log; done

Then reading log files can be done as following:

results_dir = './logs'
acc = []
for f in os.listdir(results_dir):
    with open(pjoin(results_dir, f), 'r') as fp:
        s = fp.readlines()[-1]        
    pos1 = s.find(':')
    acc.append(float(s[pos1+1:s[pos1:].find('(') + pos1]))
print(len(acc), np.mean(acc), np.std(acc))

Results

Average and std of accuracy for 10-fold cross-validation (left column). We also repeat experiments 10 times (as shown above) for different random seeds and report average and std over those 10 times (right column).

*74.72 ± 2.90 with PyTorch 1.0.0.

¹ Using hyperparameters from our paper [4] as below:

for i in $(seq 1 10); do seed=$(( ( RANDOM % 10000 ) + 1 )); python graph_unet.py -M mgcn -K 2 -f 32,32,32 --n_hidden 96 --bn --epochs 50 --lr_decay_steps 25,35,45 --lr 0.001 --seed $seed | tee logs/mcheb_proteins_"$i".log; done

Some datasets contain additional float-valued node attributes, which can improve graph classification a lot. Note that some algorithms, including Weisfeiler-Lehman (WL) Graph Kernels, are not able to make use of these additional attributes, so algorithms should be compared fairly.

These results were obtained by running (similarly for Graph U-Net, MGCN and Multigraph ChebNet):

¹ python graph_unet.py -D ENZYMES -f 128,128,128 --n_hidden 256 --lr 0.0005 --epochs 100 --lr_decay_step 150 -g

Here, lr_decay_step can be any number larger than the number of epochs to avoid learning rate decay.

² Same as above but with the -c flag.

³ Using hyperparameters from our paper [4] as below:

python graph_unet.py -D ENZYMES -M mgcn -K 4 -f 32,64,512 --n_hidden 256 --n_hidden_edge 128 --bn --lr 0.001 --epochs 50 --lr_decay_steps 25,35,45 -g

⁴ Same as above but with the -c flag.

Overvew of pooling based on Graph U-Net

Results of Graph U-Net pooling on one of the graph

Requirements

The code is tested on Ubuntu 16.04 with PyTorch 0.4.1/1.0.0 and Python 3.6.

The jupyter notebook file is kept for debugging purposes.

Optionally:

References

[1] Anonymous, Graph U-Net, submitted to ICLR 2019

[2] Cătălina Cangea, Petar Veličković, Nikola Jovanović, Thomas Kipf, Pietro Liò, Towards Sparse Hierarchical Graph Classifiers, NIPS Workshop on Relational Representation Learning, 2018

[3] Thomas N. Kipf, Max Welling, Semi-Supervised Classification with Graph Convolutional Networks, ICLR 2017

[4] Boris Knyazev, Xiao Lin, Mohamed R. Amer, Graham W. Taylor, Spectral Multigraph Networks for Discovering and Fusing Relationships in Molecules, NIPS Workshop on Machine Learning for Molecules and Materials, 2018