Installation guide

Linux and MacOSX

Dependencies

Before you proceed, make sure the following dependencies are installed:

libblas-dev
liblapack-dev

if you intend to use SuperSCS via its python interface, you also need to install

python-numpy
python-scipy

In Linux run

sudo apt-get install libblas-dev liblapack-dev

sudo apt-get install python-numpy python-scipy

Installation

First, you need to download SuperSCS from the github repo of SuperSCS, or use the command:

git clone 'https://github.com/kul-forbes/scs.git'

cd scs/

Installation runs as simple as

make

Once make finishes, the library files will be in out/.

If you want to run the tests, do

make run-test

If, additionally, you want to run the tests and perform a memory check using valgrind, do

make run-test-mem

For more advanced options, type in your terminal

make help

Docker image

Docker is the simplest and most reliable way to install SuperSCS.

The engine of SuperSCS, the C library, can be installed using the image kulforbes/superscs.

Install it running:

docker pull kulforbes/superscs

If you haven't installed docker, do so by following this guide.

You may then run the docker image and access it using an interactive terminal by running

docker run -it kulforbes/superscs

All that is necessary to run SuperSCS in C has been installed.

Compiling with SuperSCS is as simple as

gcc superscs_test.c -o superscs_test -lscsindir -lblas -llapack -lm

Interfaces

MATLAB

To install SuperSCS in MATLAB, you need to build a MEX interface.

Do:

cd matlab;

make_scs;

This should work. If not, please report an issue.

CVX MATLAB

Necessary steps:

Download and unpack CVX 3.0 from here
You need to do a tiny modification in cvx/shims/cvx_scs.m and comment-out lines 165-175. Find the modified file here.
Navigate to the cvx directory
Run cvx_setup
To use SuperSCS, we call cvx with cvx_solver scs and setting the parameter do_super_scs to 1.

Here is an example of an LP problem

A(1,1) = 0.3; A(4,1) = -0.5;
A(2,2) = 0.7; A(4,2) =  0.9; A(3,3) = 0.2;
A = sparse(A);
n = size(A,2);
b = [0.2; 0.1; -0.1; 0.1];
c = [1;-2;-3];
cvx_begin
    cvx_solver scs
    cvx_solver_settings('eps', 1e-8, 'do_super_scs', 1, 'rho_x', 1,...
         'direction', 100, 'memory', 50);
    variable x(n);
    dual variable y;
    minimize( x'*x + c' * x );
    subject to
         y : A * x <= b;
cvx_end

We have chosen the SuperSCS mode with $\rho_x=1$ , the restarted Broyden direction and memory equal to $50$ .

We have set the tolerance to $10^{-8}$ .

In case you encounter any problems, please report an issue.

Note: This a pre-alpha unstable version, so there might be issues.; For example, cvx_precision is not supported (you have to use cvx_solver_settings('eps',...)).

Python

In order to install the SuperSCS module for Python, cd to python/ and run setup.py with the argument install:

cd python/

(sudo) python setup.py install

You will then be able to import superscs into your Python code.

Further documentation for the Python module can be found here.

Table of Contents