Glue code for solving semidefinite programs in Cvxpy format using the SDPT3 package for Matlab.
Python Other
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
docs
misc
sdpt3glue
tests
.gitignore
.travis.yml
LICENSE
README.md
README.rst
requirements.in
requirements.txt
setup.py

README.md

sdpt3glue

Code Climate Build Status MIT License PyPi

The sdpt3glue package serves as glue code to allow Cvxpy semidefinite programming (SDP) problems to be solved using the SDPT3 solver for Matlab.

##Installation: This project is listed under PyPI as sdpt3glue, which means that you should be able to install it using

pip install sdpt3glue

##Documentation: Documentation is hosted here, courtesy of the Python Software Foundation.

##Quick usage guide:

Problem set up

We assume that the problem to be solved is a Cvxpy problem object representing a semidefinite program. For example, to use sdpt3glue to solve the problem

    min y s.t.
    -0.2 <= x <= -0.1
    0.4 <= z <=  0.5
    [1  x  y]
    |x  1  z| is PSD
    [y  z  1]

the following script lays the ground work by modeling the problem with Cvxpy and defining the file targets for the .mat file and the output log:

import os

from cvxpy import *
import sdpt3glue.solve as slv

# Declare variables:
X = Semidef(3)

# Define objective:
obj = Minimize(X[0, 2])

# Define constraints
constraints = [-0.2 <= X[0, 1], X[0, 1] <= -0.1, 0.4 <= X[1, 2], X[1, 2] <= 0.5]
constraints += [X[i, i] == 1 for i in range(3)]

# Construct the Cvxpy problem
problem = Problem(obj, constraints)

# Generate filenames
matfile_target = os.path.join(folder, 'matfile.mat')  # Where to save the .mat file to
output_target = os.path.join(folder, 'output.txt')    # Where to save the output log

###Problem solution four ways

1. With Matlab installed

Make sure SDPT3 is installed for Matlab and the folder containing SDPT3 is added to your MATLABPATH. Solve the problem by executing the following command:

result = sdpt3glue.sdpt3_solve_problem(P, sdpt3glue.MATLAB, matfile_target,
                                       output_target=output_target)

where matfile_target is the filepath that the problem's .mat file will be saved to and output_target is the filepath that the SDPT3 solve log will be saved to.

2. With Octave installed

Make sure SDPT3 is installed for Octave and is visible in Octave when the working folder is the folder where matfile_target is.

result = sdpt3glue.sdpt3_solve_problem(P, sdpt3glue.OCTAVE, matfile_target,
                                       output_target=output_target)

3. Using Docker

Without a locally installed copy of Matlab or Octave, but with Docker, a problem can be solved in the following way:

OCTAVE_CMD = ("docker run --rm -it -v {workdir}:/data "
              "jkawamoto/octave-sdpt3 octave").format(workdir=os.path.abspath("."))

result = sdpt3glue.sdpt3_solve_problem(P, sdpt3glue.OCTAVE, matfile_target,
                                       output_target=output_target, cmd=OCTAVE_CMD)

This makes use of a docker image created by Junpei Kawamoto for this purpose.

4. With the NEOS server

With either Firefox or phantomjs installed and for small to medium sized problems (those with .mat files of size less than about 20-25MB), the NEOS server can be used.

Disclaimer: In addition to the disclaimer provided by the MIT license, we make special note that we make no guarantees about service provided by the NEOS server, as we are not responsible for its operation. If your problem data is sensitive, take appropriate precautions.

result = sdpt3glue.sdpt3_solve_problem(P, sdpt3glue.NEOS, matfile_target,
                                       output_target=output_target)