Home

Awesome

HARP

Code for the AAAI 2018 paper "HARP: Hierarchical Representation Learning for Networks". HARP is a meta-strategy to improve several state-of-the-art network embedding algorithms, such as DeepWalk, LINE and Node2vec.

You can read the preprint of our paper on Arxiv.

This is an extended version of the original HARP with some fixes: executable under Python3 with the latest version of the imported libraries (gensim, deepwalk), refined arguments and refined magic-graph (to load correctly disconnected nodes from the MAT file). The extension is made by Artem Lutov artem@exascale.info.

Installation

The following Python packages are required to install HARP.

Magicgraph is a library for processing graph data. To install, run the following commands:

git clone https://github.com/eXascaleInfolab/magic-graph.git
cd magic-graph
# Optionally, install the library
python setup.py install

DeepWalk is an embedding learning library for graphs. To install, run the following commands:

git clone https://github.com/eXascaleInfolab/deepwalk.git
cd deepwalk
# Optionally, install the library
python setup.py install

Then, install HARP and the other requirements:

git clone https://github.com/GTmac/HARP.git
cd HARP
pip install -r requirements.txt
# Link the aforementioned libraries if they have not been installed
ln -sr ../deepwalk/deepwalk src/
ln -sr ../magic-graph/src/magicgraph src/

Usage

To run HARP on the CiteSeer dataset using LINE as the underlying network embedding model, run the following command:

python src/harp.py --input example_graphs/citeseer/citeseer.mat --model line --output citeseer.npy --sfdp-path bin/sfdp_linux

Parameters available:

--input: input_filename

  1. --format mat for a Matlab .mat file containing an adjacency matrix. By default, the variable name of the adjacency matrix is network; you can also specify it with --matfile-variable-name.

  2. --format adjlist for an adjacency list, e.g:

    1 2 3 4 5 6 7 8 9 11 12 13 14 18 20 22 32

    2 1 3 4 8 14 18 20 22 31

    3 1 2 4 8 9 10 14 28 29 33

    ...

  3. --format edgelist for an edge list, e.g:

    1 2

    1 3

    1 4

    2 5

    ...

--output: output_filename The output representations in Numpy .npy format. Note that we assume the nodes in your input file are indexed from 0 to N - 1.

--model model_name The underlying network embeddings model to use. Could be deepwalk, line or node2vec. Note that node2vec uses the default parameters, which is p=1.0 and q=1.0.

--sfdp-path sfdp_path Path to the binary file of SFDP, which is the module we used for graph coarsening. You can set it to sfdp_linux, sfdp_osx or sfdp_windows.exe depending on your operating system.

--workers procs_num=cpu_num The number of parallel executors, equals to the number of logical CPUs by default.

More options: The full list of command line options is available with python src/harp.py --help.

Evaluation

To evaluate the embeddings on a multi-label classification task, run the following command:

python src/scoring.py -e citeseer.npy -i example_graphs/citeseer/citeseer.mat -t 1 2 3 4 5 6 7 8 9

Where -e specifies the embeddings file, -i specifies the .mat file containing node labels, and -t specifies the list of training example ratios to use.

Note

SFDP is a library for multi-level graph drawing, which is a part of GraphViz. We use SFDP for graph coarsening in this implementation. Note that SFDP is included as a binary file under /bin; please choose the proper binary file according to your operation system. Currently we have the binary files under OSX, Linux and Windows.

Citation

If you find HARP useful in your research, please cite our paper:

@inproceedings{harp,
	title={HARP: Hierarchical Representation Learning for Networks},
	author={Chen, Haochen and Perozzi, Bryan and Hu, Yifan and Skiena, Steven},
	booktitle={Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence},
	year={2018},
	organization={AAAI Press}
}