subgraph2vec

This repository contains the "tensorflow" implementation of our paper "subgraph2vec: Learning Distributed Representations of Rooted Sub-graphs from Large Graphs". The paper could be found at: https://arxiv.org/pdf/1606.08928.pdf

Dependencies

This code is developed in python 2.7. It is ran and tested on Ubuntu 14.04 and 16.04. It uses the following python packages:

1. tensorflow (version == 0.12.1)
2. networkx (version <= 1.11)
3. joblib (version <= 0.11)
4. scikit-learn (+scipy, +numpy)

The procedure for setting up subgraph2vec is as follows:

1. git clone the repository (command: git clone https://github.com/MLDroid/subgraph2vec_tf.git )
2. untar the data.tar.gz tarball

The procedure for obtaining rooted subgraph vectors using subgraph2vec and performing graph classification is as follows:

1. move to the folder "src" (command: cd src) (also make sure that kdd 2015 paper's (Deep Graph Kernels) datasets are available in '../data/kdd_datasets/dir_graphs/')
2. run main.py --corpus <dataset of graph files> --class_labels_file_name <file containing class labels of graphs to be used for graph classification> file to:
	*Generate the weisfeiler-lehman kernel's rooted subgraphs from all the graphs 
	*Train skipgram model to learn subgraph embeddings. The same will be dumped in ../embeddings/ folder
	*Perform graph classification using graph kernel and deep graph kernel
3. example: 
	*python main.py --corpus ../data/kdd_datasets/mutag --class_labels_file_name ../data/kdd_datasets/mutag.Labels 
	*python main.py --corpus ../data/kdd_datasets/proteins --class_labels_file_name ../data/kdd_datasets/proteins.Labels --batch_size 16 --embedding_size 128 --num_negsample 5

Other command line args:

optional arguments:
	-h, --help            show this help message and exit
	-c CORPUS, --corpus CORPUS
			        Path to directory containing graph files to be used
			        for graph classification or clustering
	-l CLASS_LABELS_FILE_NAME, --class_labels_file_name CLASS_LABELS_FILE_NAME
			        File name containg the name of the sample and the
			        class labels
	-o OUTPUT_DIR, --output_dir OUTPUT_DIR
			        Path to directory for storing output embeddings
	-b BATCH_SIZE, --batch_size BATCH_SIZE
			        Number of samples per training batch
	-e EPOCHS, --epochs EPOCHS
			        Number of iterations the whole dataset of graphs is
			        traversed
	-d EMBEDDING_SIZE, --embedding_size EMBEDDING_SIZE
			        Intended subgraph embedding size to be learnt
	-neg NUM_NEGSAMPLE, --num_negsample NUM_NEGSAMPLE
			        Number of negative samples to be used for training
	-lr LEARNING_RATE, --learning_rate LEARNING_RATE
			        Learning rate to optimize the loss function
	--n_cpus N_CPUS       Maximum no. of cpu cores to be used for WL kernel
			        feature extraction from graphs
	--wlk_h WLK_H         Height of WL kernel (i.e., degree of rooted subgraph
			        features to be considered for representation learning)
	-lf LABEL_FILED_NAME, --label_filed_name LABEL_FILED_NAME
			        Label field to be used for coloring nodes in graphs
			        using WL kenrel
	-v VALID_SIZE, --valid_size VALID_SIZE
			        Number of samples to validate training process from
			        time to time

Contact

In case of queries, please email: annamala002@e.ntu.edu.sg OR XZHANG048@e.ntu.edu.sg

Reference

Please consider citing the follow paper when you use this code.
@article{narayanansubgraph2vec,
  title={subgraph2vec: Learning Distributed Representations of Rooted Sub-graphs from Large Graphs},
  author={Narayanan, Annamalai and Chandramohan, Mahinthan and Chen, Lihui and Liu, Yang and Saminathan, Santhoshkumar}
}

Acknowledgements

Thanks to Zhang Xinyi (https://github.com/XinyiZ001) for the support on coding/testing subgraph2vec TF version.