We propose a method for characterizing large complex networks by introducing a new matrix structure, unique for a given network, which encodes structural information; provides useful visualization, even for very large networks; and allows for rigorous statistical comparison between networks. Dynamic processes such as percolation can be visualized using animation.
B_matrix.py and B_matrix.cpp take an edgelist file and write the
corresponding B-matrix to a file
An edgelist is an M x 2 matrix for a graph with M edges. The C++ code requires nodes be sequential integers numbered from zero, while the python code is slower, but much more flexible (can handle directed networks for example, which the C++ cannot) and forgiving. Python code requires networkx and (optionally) pylab to plot.
Unless the networks are very large, I greatly encourage using the python code
instead of the C++. Additionally, networkx has undergone a great deal of
updates and the B_matrices.py file might be slightly out of date, requiring
small changes to work again. Fair warning, buyer beware, etc.
Various matlab m-files are included for loading a B-matrix from file,
trimming empty columns, and computing the distance between two matrices. The
latter is accomplished using B_Distance.m, which takes two matrices as input
and returns the distance as described in the paper. It will optionally also
plot the row-wise distances.
The m-files may also work in octave, an open source "clone" of matlab but I haven't tried so I make no guarantees.
- Jim Bagrow, 2008-04-17
- bagrowjp [at] gmail [dot] com