Implementation of Single-level and Multilevel sparsification. See paper here: http://arxiv.org/abs/1601.05527

Note: This project is still in development.

To build:

make

To run:

For single-level: ./msparse -i ~/test_graphs/fb-uf.edges -n 35111 --zero-index --gname fb-uf -o out.edges -s 0.4 --single-level --weak

For multi-level ./msparse -i ~/test_graphs/fb-uf.edges -n 35111 --zero-index --gname fb-uf -o out.edges --param-list 0.1,0.1,0.1 --weak

To set number of openmp threads (set to 8 by default) export OMP_NUM_THREADS=16

To disable openmp, remove '-fopenmp' from LFLAGS in Makefile

Notes:

==> Param list starts from coarsest level

==> c++11 required (version g++ 4.8 or over)

==> Graph file must already be undirected, node numbering has to be continuous and need only contain one direction of relationship e.g

Good graph file
	1 2
	3 4
	5 3

Bad graph file
	1 2
	2 1
	3 4
	4 3
	5 3
	3 5
graph files (edge list) can also contain 3rd column for weights.

==> Space delimeter is used by default. To change to tab use --delim tab

==> Weak + Strong edges sparsified by default.

Options

--weak: retain weak edges --strong: Retain strong edges --normalize: normalize algebraic distance by product of degrees.

-i - path of input graph -n - number of nodes --gname - shortname for graph. --zero-index - index of first node starts at 0. --permute - random shuffle node ordering -r - interpolation order. Default 1. --delim - delimeter of input file. Only space or tab allowed.

See Main.cpp for all available options.

Output:

sparsified graph

emmanuj/ml-sparsifier

Notes:

Options