mvgrl

GammaGL Implementation of MVGRL

This GammaGL example implements the model proposed in the paper Contrastive Multi-View Representation Learning on Graphs.

Author's code: https://github.com/kavehhassani/mvgrl

Datasets

Unsupervised Graph Classification Datasets:

'MUTAG', 'PTC_MR', 'REDDIT-BINARY', 'IMDB-BINARY', 'IMDB-MULTI'.

Dataset	MUTAG	PTC_MR	RDT-B	IMDB-B	IMDB-M
# Graphs	188	344	2000	1000	1500
# Classes	2	2	2	2	3
Avg. Graph Size	17.93	14.29	429.63	19.77	13.00

• RDT-B, IMDB-B, IMDB-M are short for REDDIT-BINARY, IMDB-BINARY and IMDB-MULTI respectively.

Unsupervised Node Classification Datasets:

'Cora', 'Citeseer' and 'Pubmed'

Dataset	# Nodes	# Edges	# Classes
Cora	2,708	10,556	7
Citeseer	3,327	9,228	6
Pubmed	19,717	88,651	3

Arguments

Graph Classification:

--dataset         str     The graph dataset name.                Default is 'MUTAG'.
--n_epochs          int     Number of training periods.            Default is 200.
--patience         int     Early stopping steps.                  Default is 20.
--lr               float   Learning rate.                         Default is 0.001.
--l2_coef               float   Weight decay.                          Default is 0.0.
--batch_size       int     Size of a training batch.              Default is 64.
--num_layers         int     Number of GNN layers.                  Default is 4.
--hidden_dim          int     Embedding dimension.                   Default is 32.

Node Classification:

--dataset         str     The graph dataset name.                Default is 'cora'.
--n_epochs          int     Number of training periods.            Default is 500.
--patience         int     Early stopping steps.                  Default is 20.
--lr             float   Learning rate of main model.           Default is 0.001.
--clf_lr             float   Learning rate of linear classifer.     Default is 0.01.
--l2_coef              float   Weight decay of main model.            Default is 0.0.
--clf_l2_coef              float   Weight decay of linear classifier.     Default is 0.0.
--epsilon          float   Edge mask threshold.                   Default is 0.01.
--hidden_dim          int     Embedding dimension.                   Default is 512.
--sample_size      int     Subgraph size.                         Default is 2000.

How to run examples

Graph Classification

# Enter the 'graph' directory
cd graph

# MUTAG:
TL_BACKEND=paddle python mvgrl_trainer.py --dataset MUTAG --n_epoch 20 --hidden_dim 32
TL_BACKEND=tensorflow python mvgrl_trainer.py --dataset MUTAG --n_epoch 20 --hidden_dim 32
TL_BACKEND=torch python mvgrl_trainer.py --dataset MUTAG --n_epoch 20 --hidden_dim 32

# PTC_MR:
TL_BACKEND=paddle python mvgrl_trainer.py --dataset PTC_MR --epochs 32 --hidden_dim 128
TL_BACKEND=tensorflow python mvgrl_trainer.py --dataset PTC_MR --epochs 32 --hidden_dim 128
TL_BACKEND=torch python mvgrl_trainer.py --dataset PTC_MR --epochs 32 --hidden_dim 128

# IMDB-BINARY

TL_BACKEND=paddle  python mvgrl_trainer.py --dataset IMDB-BINARY --epochs 20 --hidden_dim 512 --n_layers 2
TL_BACKEND=tensorflow python mvgrl_trainer.py --dataset IMDB-BINARY --epochs 20 --hidden_dim 512 --n_layers 2
TL_BACKEND=torch python mvgrl_trainer.py --dataset IMDB-BINARY --epochs 20 --hidden_dim 512 --n_layers 2
# IMDB-MULTI
TL_BACKEND=paddle python mvgrl_trainer.py --dataset IMDB-MULTI --epochs 20 --hidden_dim 512 --n_layers 2
TL_BACKEND=tensorflow python mvgrl_trainer.py --dataset IMDB-MULTI --epochs 20 --hidden_dim 512 --n_layers 2
TL_BACKEND=torch python mvgrl_trainer.py --dataset IMDB-MULTI --epochs 20 --hidden_dim 512 --n_layers 2

Node Classification

For semi-supervised node classification on 'Cora', 'Citeseer'

full-graph training, see 'mvgrl_trainer.py', where we contrast the local and global representations of the whole graph.

# Enter the 'node' directory
cd node

# Cora with full graph
TL_BACKEND=paddle python mvgrl_trainer.py --dataset cora --lr 0.0005 --patience 40
TL_BACKEND=tensorflow python mvgrl_trainer.py --dataset cora 
TL_BACKEND=torch python mvgrl_trainer.py --dataset cora 

# Citeseer 
TL_BACKEND=paddle python mvgrl_trainer.py --dataset citeseer
TL_BACKEND=tensorflow python mvgrl_trainer.py --dataset citeseer
TL_BACKEND=torch python mvgrl_trainer.py --dataset citeseer

Performance

We use the same hyper-parameter settings as stated in the original paper.

Graph classification:

Dataset	MUTAG	PTC-MR	REDDIT-B	IMDB-B	IMDB-M
Accuracy Reported	89.7	62.5	84.5	74.2	51.2
DGL	89.4	62.2	85.0	73.8	51.1
GammaGL(tf)	89.51 ± 0.62	60.25 ± 0.60	OOM	73.28 ± 0.61	50.5 ± 0.23
GammaGL(th)	89.30 ± 0.71	60.50 ± 1.95	OOM	73.16 ± 0.42	50.95 ± 0.56
GammaGL(pd)	88.39 ± 0.68	60.32 ± 1.96	OOM	72.92 ± 0.46	51.27 ± 0.57
GammaGL(ms)	--.-	--.-	--.-	--.-	--.-

• The datasets that the authors used are slightly different from standard TUDataset (see gammagl.data.TUDataset) in the nodes' features(e.g. The node features of 'MUTAG' dataset are of dimensionality 11 rather than 7)

Node classification:

Dataset	Cora	Citeseer	Pubmed
Accuracy Reported	86.8	73.3	80.1
DGL-sample	83.2	72.6	79.8
DGL-full	83.5	73.7	OOM
GammaGL(tf)	82.73 ± 0.33	70.96 ± 0.43	OOM
GammaGL(th)	81.71 ± 0.85	70.9 ± 0.56	OOM
GammaGL(pd)	82.23 ± 1.09	70.91 ± 0.06	OOM
GammaGL(ms)	--.-	--.-	OOM

• We fail to reproduce the reported accuracy on 'Cora', even with the authors' code.
• The accuracy reported by the original paper is based on fixed-sized subgraph-training.
• The model performance is the average of 5 tests