The bvgraph class makes working with enormous graph adjacency matrices in Matlab a snap. The class implements an interface to BVGraph files (http://webgraph.dsi.unimi.it) that emulates a Matlab sparse matrix.
>> G = bvgraph(‘data/wb-cs.stanford’);
>> y = G*rand(size(G,1),1);
are all valid operations. See the “bvgraph example” file for a more exhaustive explanation.
The goal of the library is to support computing PageRank with a 1 billion link graph and 120 million nodes possible inside Matlab with only 2 GB of memory. To load the matrix itself on a 64-bit platform would take around 10 GB of memory. As the “huge graphs” example shows, the bvgraph class achieves this goal.
To find additional datasets, see http://law.dsi.unimi.it/index.php?option=com_include&Itemid=65
The software was tested on
Matlab 2007b on amd64 Ubuntu 7.04
Matlab 2007a on amd64 Ubuntu 7.04
Matlab 2006b on amd64 Ubuntu 7.04
Matlab R14SP3 on qemu (i686) Ubuntu 6.06.1
Matlab 2007b on i686 Ubunutu 5.10
Matlab 7.0 on i386 Windows
Matlab 2007a on i386 Windows
Matlab 2007a on qemu (x86_64) WinXP64
Removed GPLed pieces.
New version 1.1 with bug fixes for 32-bit matlab and now precompiled for 64-bit windows.
Fixed two packaging errors in the zip file. Removed pagerank.m and duplicate of huge_graph.html.