node2vec-c

node2vec implementation in dependency-less C++. node2vec uses short biased random walks to learn representations for vertices in unweighted graphs. Other implementations are available in C++ in SNAP project and a reference one in Python + Gensim.

This code was developed to be used in out paper, VERSE for a fair performance comparison. Moreover, the high-performance implementation in SNAP did not match the performance of the original implementation (as for 2017).

Installation and usage

For the executable:

make

should be enough on most platforms. If you need to change the default compiler (i.e. to Intel), use:

make CXX=icpc

Intel® FMA availability is crucial for performance of the implementation, meaning the processor Haswell (2013). You will get a warning on runtime if your processor does not support it.

Usage

Usage: node2vec [OPTIONS]

Options:
  -input PATH                    Input file in binary CSR format
  -output PATH                   Output file, written in binary
  -threads INT                   Number of threads to use (default 1)
                                   Note: hyperthreading helps as well
  -dim INT                       node2vec parameter d: dimensionality of
                                   embeddings (default 128)
  -nwalks INT                    node2vec parameter gamma: number of walks per
                                   node (default 80)
  -walklen INT                   node2vec parameter t: length of random walk
                                   from each node(default 80)
  -window INT                    node2vec parameter w: window size (default 10)
  -nsamples INT                  node2vec parameter k: number of negative samples (default 5)
  -p FLOAT                       node2vec parameter p: random walk bias (default 1)
  -q FLOAT                       node2vec parameter q: random walk bias (default 1)
  -lr FLOAT                      Initial learning rate
  -seed INT                      Sets the random number generator seed to INT
  -verbose INT                   Controls verbosity level in [0,1,2], 0 meaning
                                   nothing will be displayed, and 2 mening
                                   training progress will be displayed.

Graph format

This implementation uses a custom graph format, namely binary compressed sparse row (BCSR) format for efficiency and reduced memory usage. Converter for three common graph formats (MATLAB sparse matrix, adjacency list, edge list) can be found in the root directory of our main code repository.

Citing

If you find node2vec useful in your research, we ask that you cite the original paper:

@inproceedings{Grover:2016:NSF:2939672.2939754,
    author = {Grover, Aditya and Leskovec, Jure},
    title = {Node2Vec: Scalable Feature Learning for Networks},
    booktitle = {Proceedings of the 22Nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
    series = {KDD '16},
    year = {2016},
    isbn = {978-1-4503-4232-2},
    location = {San Francisco, California, USA},
    pages = {855--864},
    numpages = {10},
    url = {http://doi.acm.org/10.1145/2939672.2939754},
    doi = {10.1145/2939672.2939754},
    acmid = {2939754},
    publisher = {ACM},
    address = {New York, NY, USA},
    keywords = {feature learning, graph representations, information networks, node embeddings},
}

Contact