nishi951 / cvxbenchmarks

A testing framework for convex solvers in CVXPY.

cvxbenchmarks

A testing framework for convex solvers in CVXPY.

Installation

After cloning the repo, run:

$ cd cvxbenchmarks
$ python setup.py install

If you have pytest installed (e.g. by running pip install pytest) You can check your installation by running:

$ py.test

Example

cvxpy comes with the ECOS and SCS solvers already installed. To see how they compare on some small sample problems, navigate to the cvxbenchmarks directory and run

$ cvxbench generate --example
$ cvxbench run --example

A pandas dataframe with relevant timing and size information should appear, and a figs.pdf file containing some useful plots should also be generated.

Problems

The cvxbenchmarks problem database has a wide variety of parametrizable problem types. To see the types of problems available, check out the cvxbenchmarks/examples directory, which contains a variety of html files generated using ipython nbviewer. The original ipython notebooks can be found in cvxbenchmarks/examples/ipython.

In addition to the cvxbenchmarks problem database, it is possible to download problems from a number of other places for use with cvxbenchmarks.

Currently supported: qpoases

In the works: - CUTest - DIMACS - CVXCanon

Adding your own problems

TODO

Cvxbenchmarks was designed and implemented by Mark Nishimura with the input of Steven Diamond and Stephen Boyd.