horacepan / MLGkernel

Implementation of the Multiscale Laplacian Graph Kernel

The Multiscale Laplacian Graph Kernel

This is a C++ implementation of the Multiscale Graph Laplacian kernel as described in:

R. Kondor, H. Pan, The Multiscale Graph Laplacian (2016)

Requirements

• C++11
• Eigen

Installation/Setup

Change the EIGENDIR variable Makefile.options to the path to your installation of the Eigen library. Run the following command to create the runMLG executable in the MLGkernel directory.

make all

Run the demo

After compiling, you can run the sample.sh script, which will run the MLGkernel on the sample datasets in the data directory and save the resulting kernel matrices in data/results.

sh sample.sh

Content

To understand the code, the main files worth skimming through are in the MLGkernel directory: MLGdataset, Linearizer, FLGinstance, and FLGkernel. See MLGkernel/runMLG.cpp for an example of how to load data/feature files and how to compute/save the MLG kernel matrix.

Data Format

In order to use your own data, you have to provide the data via

• a text file in the following format: the first line is the number of graphs in the dataset The next line is the number of vertices in first graph, the next {size of first graph} lines is the space delimited adjacency matrix of the first graph. And so on for the rest of the graphs. See data/MUTAG.txt for an example of this format.
• a text file containing a discrete node label of each graph for each graph in the dataset. The first line is the number of graphs in the dataset. The second line is the size of the first graph in the dataset. The next {size of first graph} lines denote the node labels of the vertices of the first graph. And so on for the rest of the graphs. See data/MUTAG_nodelabels.txt for an example of this data format.

Note that if no feature file is given for the runMLG executable, it will default to using vertex degrees as node labels.

Contact

If you have any questions/comments/concerns, please contact hopan[at]uchicago.edu. In particular, it would be fantastic if you could report any bugs!