GitHub - arbenson/tensor-sc: Tensor spectral clustering. (original) (raw)

tensor-sc

Tensor Spectral Clustering. This is the code used in the paper:

Austin R. Benson, David F. Gleich, and Jure Leskovec. "Tensor Spectral Clustering for Partitioning Higher-order Network Structures". In Proceedings of the 2015 SIAM International Conference on Data Mining (SDM), 2015.

The full version of the paper is available here.

The code is released under the simplified BSD license. Please see the LICENSE file.

For help using the code, please contact Austin: arbenson AT stanford DOT edu.

Getting started

What you need:

• Compiler supporting C++11
• SNAP for C++

Edit the Makefile for your compiler and the location of SNAP.

Layered flow example

First, build the example:

Run the algorithms:

Look at the communities for tensor spectral clustering, the directed Laplacian, and the subgraph directed Laplacian:

cat layered_flow_tsc_comms.txt
cat layered_flow_dl_comms.txt
cat layered_flow_subdl_comms.txt

Anomaly detection example

First, build the example:

Run the algorithms:

Look at the communities for tensor spectral clustering, the directed Laplacian, and the subgraph directed Laplacian:

cat anomaly_tsc_comms.txt
cat anomaly_dl_comms.txt
cat anomaly_subdl_comms.txt

Big network directed 3-cycle cut example

We will now go through the steps for running the directed 3-cycle cut algorithms on wiki-Vote, one of the SNAP networks. First, download the data:

wget http://snap.stanford.edu/data/wiki-Vote.txt.gz
gunzip wiki-Vote.txt.gz
mv wiki-Vote.txt data/

Now, filter the graph:

bash scripts/filter_data.sh wiki-Vote

Build the partitioning code:

Run the algorithm:

Visualize the number of directed 3-cycles cut as a function of the smaller partition size:

python scripts/d3c_plot.py wiki-Vote-filter num_cut