ABOUT
-----

Demonstrates the usage of the LPQP solver for the Potts model. The actual LPQP
code can be obtained from:

https://github.com/ppletscher/lpqp

The code was slightly modified after the initial ICML publication, but the results
seem to be pretty much the same.


INSTALL
-------

The experiments are done using matluster, a collection of Matlab scripts which
simplifies the experimental evaluations on a cluster. Download it to a
subdirectory called 'matluster' using the following command:

git clone https://github.com/ppletscher/matluster

Then you can run the experiments by following these steps. This works on the
cluster of ETH Zurich, for different systems you might need to change a few
things.

1. copy algo_triplet binary for mplp to grid/ directory.
2. link to mplp and lpqp matlab code.
3. run ./compile_on_brutus.sh .
4. in Matlab run grid_experiment_prepare .
5. on command line submit the jobs using grid_experiment.sh .
6. collect all the results by running grid_experiment_collect in Matlab.