Python interface for SCS 2.0.0 and higher.

To install using pip (recommended) use:

pip install scs

To install SCS from source:

git clone --recursive https://github.com/bodono/scs-python.git
cd scs-python
python setup.py install

You may need sudo privileges for a global installation. Running SCS requires numpy and scipy to be installed. You can install the gpu interface using

python setup.py install --scs --gpu

To test that SCS installed correctly and you have nose installed, run

nosetests

you can also solve a random cone problem (without nose) using

python test/solve_random_cone_prob.py

After installing the SCS interface, you import SCS using

import scs

This module provides a single function solve with the following call signature:

sol = scs.solve(data,
                cone,
                use_indirect=False,
                gpu=False,
                verbose=True,
                normalize=True,
                max_iters=5000,
                scale=1,
                eps=1e-5,
                cg_rate=2,
                alpha=1.5,
                rho_x=1e-3,
                acceleration_lookback=10,
                write_data_filename=None)

The argument data is a python dictionary with three elements A, b, and c where b and c are 1d NUMPY arrays and A is a SCIPY sparse matrix in CSC format; if they are not of the proper format, SCS will attempt to convert them.

The argument cone is a dictionary with fields f, l, q, s, ep, ed, and p (all of which are optional) corresponding to the supported cone types.

The returned object is a dictionary containing the keys 'x', 'y', 's', and 'info'. The first three are NUMPY arrays containing the respective solution vector. The 'info' value is a dictionary with solver information.