subgraph2vec

The subgraph2vec paper could be found at: https://arxiv.org/pdf/1606.08928.pdf

Virtual Machine with working subgraph2vec setup

• An Ubuntu 16.04 vitural machine with all the libraries required for running subgraph2vec could be downloaded from: https://drive.google.com/open?id=0B5UPYObHDu7iWEE0ZlV5WXlhOGs
• Credentials of the VM: username = sg2vec, password = sg2vec
• Just clone this repository and follow the instructions mentioned below to run subgraph2vec on this VM
• NOTE: This VM contains a older version of this repository in its home folder. So, please delete it and clone afresh for accessing the latest code.

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. gensim (version == 0.12.4)
2. networkx (version <= 1.11)
3. joblib (version <= 0.10.3)
4. scikit-learn (+scipy, +numpy)

Subgraph2vec is developed on top of "gensim" python package. We have edited the native word2vec_inner.py file and created a new .so file named 'mod_word2vec_inner.so'. NOTE: tensorflow version of the same will be released soon.

The procedure for setting up subgraph2vec is as follows:

1. git clone the repository (command: git clone https://github.com/MLDroid/subgraph2vec.git )
2. sudo run init.sh (installs gensim, modifies word2vec libraries to use radial skipgram (refer sec: 5.2.2 of the paper))
NOTE: no need to run init.sh on the sg2vec VM (it is already setup)

The procedure for verifying the setup is as follows:

1. make sure the gensim package contains a softlink "word2vec_inner.so -> mod_word2vec_inner.so" (command: ls -l /usr/local/lib/python2.7/dist-packages/gensim/models)
2. make sure that the file named 'Annaword2vec.py' is present in /usr/local/lib/python2.7/dist-packages/gensim/models (command: ls /usr/local/lib/python2.7/dist-packages/gensim/models/Annaword2vec.py)
3. make sure that the version of gensim is 0.12.4 (command from the python prompt: import gensim;print gensim.__version__)

The procedure for obtaining rooted subgraph vectors using subgraph2vec approach is as follows:

1. move to the folder "src-code" (command: cd src_code) (also make sure that kdd 2015 paper's (Deep Graph Kernels) datasets are available in '../kdd_datasets/dir_graphs/')
2. run dump_wl_kernel_sentences.py file to generate the weisfeiler-lehman kernel's rooted subgraphs from all the graphs in a given folder
   * syntax: python dump_wl_kernel_sentences.py <gexf/json graph_dir> <height of WL kernel> <num of cpu cores for multi-processing>
   * example: python dump_wl_kernel_sentences.py ../kdd_datasets/dir_graphs/mutag 4 8 (this will generate files with extension .WL4 that contain target and context subgraphs in every graph)
3. run sg2vec.py to generate embeddings for each of the subgraphs generated in step 2
   * syntax: python sg2vec.py <src_dir> <file_extension> <opfname_prefix> <embedding_dim> <iterations> <# of cpu cores>
   * example: python sg2vec.py ../kdd_datasets/dir_graphs/mutag WL4 mutag 32 100 8
   * this will generate the following two outputs:
      * i)  a gensim word2vec model containing embeddings for all the subgraphs in the "../models" folder
      * ii) a json dump of the python dictionary containg embeddings of subgraphs (format of the dictionary: key = subgraph, value = embedding)

Contact

In case of queries, please email: [email protected]

Reference

@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}
}