Author: Gopal Yalla and Tim Smith
Build Date: 10/08/2018
The SPARC package is an educational tool for solving the sparse page rank problem using spectral bisection, graph coarsening, and compact sparse column (CSC) format. To aid in documentation, usage, and installation, please see the related pages:
Bugs in the code and errors or omissions in the documentation can be reported to gopal@ices.utexas.edu or timsmith204@utexas.edu. Requests and contributions are welcome at the same e-mail addresses.
Below are results obtained using SPARC.
The figure on the left is original data from facebook that has 88234 non-zero entries and 16466 number of non-zero entries on the off diagonal. The figure on the left is the same data after spectral bisection. The number of off diagonal entries reduces to 1266.
The figure on the left is original DBLP data that has 1049866 non-zero entries and 628800 number of non-zero entries on the off diagonal. The figure on the left is the same data after spectral bisection. The number of off diagonal entries reduces to 350038.
The figure on the left is the same original DBLP data as above with 1049866 non-zero entries and 628800 number of non-zero entries on the off diagonal. The figure on the left is the data after four levels of coarsening followed by spectral bisection. The number of off diagonal entries reduces to 523194.
Lastely, the figure below shows initial data undergoing three levels of coarsening and then spectral bisection followed by reordering. The number of off diagonal non zeros reduced from 830 to 482.
