GraphSim

This is the repo for Learning-based Efficient Graph Similarity Computation via Multi-Scale Convolutional Set Matching (AAAI 2020), and Convolutional Set Matching for Graph Similarity. (NeurIPS 2018 Relational Representation Learning Workshop).

Data and Files

Get the data files <dataset>_result.zip and extract under data.

Get the pickle files <dataset>_<ged_or_mcs>_<algo>_gidpair_dist_map.pickle and put under save.

Get the (ground-truth and baseline) result files <dataset>_result.tar.gz and extract under result.

This repo only contains code and please download the above files from https://drive.google.com/drive/folders/1JcAgWKYC41687UeiLaFg-QlPmIpZvWhT?usp=sharing

Dependencies

Install the following the tools and packages:

• python3: Assume python3 by default (use pip3 to install packages).
• numpy
• pandas
• scipy
• scikit-learn
• tensorflow (1.8.0 recommended)
• networkx==1.10 (NOT 2.1)
• beautifulsoup4
• lxml
• matplotlib
• seaborn
• colour
• pytz
• requests
• klepto
• pygraphviz. The following is an example set of installation commands (tested on Ubuntu 16.04)

  sudo apt-get install graphviz libgraphviz-dev pkg-config
  pip3 install pygraphviz --install-option="--include-path=/usr/include/graphviz" --install-option="--library-path=/usr/lib/graphviz/"
  

• Graph Edit Distance (GED):
  • graph-matching-toolkit
    • Need java
    • Follow the instructions on https://github.com/dan-zam/graph-matching-toolkit to compile
  • F2, F2LP, F24threads
    • Obtain from https://drive.google.com/file/d/12MBjXcNko83mAUGKe9nVJqEKjLTjDJNd/view?usp=sharing
    • Put under /model/<F2/F2LP/F24threads>/
  • hed
    • Obtain from https://github.com/priba/aproximated_ged
    • Put under /model/aproximated_ged/
• Maximum Common Subgraph (MCS):
  • mccreesh2017
    • Obtain it through running cd src && git clone https://github.com/yunshengb/mccreesh2017.git
    • Put under /model/mcs/
    • Need g++ compiler
    • Compile the binary called mcsp in model/mcs/mccreesh2017/code/james-cpp and put the binary in model/mcs/mccreesh2017

Reference commands: sudo pip3 install numpy pandas scipy scikit-learn tensorflow networkx==1.10 beautifulsoup4 lxml matplotlib seaborn colour pytz requests klepto

Tips for PyCharm Users

• If you see red lines under import, mark src and model/Siamese as Source Root, so that PyCharm can find those files.
• Mark model/Siamese/logs and model/Siamese/exp as Excluded, so that PyCharm won't spend time inspecting those logs.

Run

Modify the configuration file model/Siamese/config.py, then run the model.

Example commands to run our model:

cd model/Siamese
python3.5 run.py 

The model's results are saved under model/Siamese/logs. To check prec@k, etc., use model/Siamese/extract_prec.py.

How to create your own datasets?

Define your data object in src/data.py and src/utils.py.

Run the GED/MCS solver(s) on your datasets using src/exp.py (exp1). Notice that you need to set up the GED/MCS solver(s) according to the Dependencies above. The results are saved under result/.

Load your dataset into the model by modifying model/Siamese/config.py and run the model as described above.