snap-python

1. Install SWIG for your platform (see below). Swig should be able to run from the command-line.

2. Checkout the snap-python repository as well as the SNAP C++ repository.

    git clone git@github.com:snap-stanford/snap-python.git
    git clone git@github.com:snap-stanford/snap.git
   

3. Then, run make from the top-level of snap-python. This will make the SNAP code into a Python module, using SWIG. Finally, it will run some Python tests in the test directory.

    cd snap-python
    make
   

   From a Python interpreter, you should be able to import snap module:

    $ python
    >>> import sys
    >>> sys.path.append("swig")
    >>> import snap
   

4. There are some examples in the examples directory. For example, to run benchmarks:

    $ cd examples
    $ python benchmark.py -h
    usage: benchmark.py [-h] [-v] [-r RANGE] [-e EDGES_DEG] [-d] [-t GRAPH_TYPES]
                        [-n NUM_ITERATIONS] [-o OUTPUT_FILE] [-g] [-w]
   
    optional arguments:
      -h, --help            show this help message and exit
      -v, --verbose         increase output verbosity
      -r RANGE, --range RANGE
                            range (4-6) (10^4 to 10^6 nodes)
      -e EDGES_DEG, --edges_deg EDGES_DEG
                            range of degrees (e.g "2-3" => (10^1 to 10^3 edges per
                            node)
      -d, --deterministic   deterministic benchmark
      -t GRAPH_TYPES, --graph_types GRAPH_TYPES
                            Graph types, comma separated. Available: rand_ungraph,
                            rand_ngraph, rmat, pref, sw
      -n NUM_ITERATIONS, --num_iterations NUM_ITERATIONS
                            number of iterations
      -o OUTPUT_FILE, --output_file OUTPUT_FILE
                            file to output results
      -g, --generate        generate new graphs
      -w, --write_graph     save graph
    $ python benchmark.py -v -g -r 4-6	# needs about 4.3GB RAM and 4 min to run
   

SWIG Installation

Linux

Follow the instructions from SWIG's website: download, configure and make, SWIG files. Or, use your built-in installer (a CentOS example):

sudo yum install swig

Mac OS X

swig-1.3.12 and later support OS-X/Darwin.

0. If you have homebrew, simply hit brew install swig in terminal and ignore the rest of the instructions. Otherwise, download the Unix sources, configure, and build from the command terminal. This has been tested on 10.8.2. The following is adopted from ColourBlomb.

1. Download the Unix source from http://swig.org/download.html

2. Moving to the terminal, extract the files from the tarball and move to the root directory of the SWIG install:

    cd /Developer/SWIG
    tar -xf swig-2.0.4.tar.gz
    cd swig-2.0.4
   

3. Run ./configure. This will produce an error if you don't have the PCRE (Perl Compatible Regular Expressions) library package installed. This dependency is needed for configure to complete. Either:

   • Install the PCRE developer package on your system (preferred approach).

   • Download the PCRE source tarball, build and install on your system as you would for any package built from source distribution.

   • Use the Tools/pcre-build.sh script to build PCRE just for SWIG to statically link against. Run Tools/pcre-build.sh –help for instructions. (quite easy and does not require privileges to install PCRE on your system)

   • Configure using the –without-pcre option to disable regular expressions support in SWIG (not recommended). See config.log for more details.

       make
       sudo make install
     

4. PCRE should now have successfully installed so move to the swig install directory and try ./configure again:

    cd ../swig-2.0.4
    ./configure
   

   This time no errors are thrown so try and install:

    make
    sudo make install
   

5. Once this has completed test that SWIG has installed correctly, type swig into the terminal and hopefully you’ll get the response: Must specify an input file. Use -help for available options.

SWIG Benchmarks

Example SWIG programs using the SNAP Ringo for multi-attribute edges are in the examples directory. The benchmark program benchmark.py performs a series of functions on the graph data, including node/edge iteration, degree checks, clustering coefficients, largest weakly and strongest components, etc. For R-MAT graphs with 1 million nodes and 10 million edges, this takes on average:

• On CentOS 6.3 with 2.66 GHz processor, 19.71 sec to generate a new graph and and 17.49 sec to run the tests.
• On Mac OSX 10.8 with 2.6 GHz processor, 13.95 sec to generate and 15.06 sec to run the tests.

To run a benchmark test you can run the following command:

python benchmark.py --verbose -n 5 --range 4-7 --type rmat --generate