Benchmarking against other differentiable convex optimization softwares

[akshay326]

Although not now, some day down the road it might be good to benchmark our code (in terms of speed / supported problem types) against other existing solutions. Some possible candidates:

• https://github.com/cvxgrp/cvxpylayers
• https://github.com/locuslab/qpth

[akshay326]

how could i forget :sweat_smile:

• https://github.com/cvxgrp/diffcp

[matbesancon]

yes agreed, setting up benchmark can be quite some work, a good starting point will be setting up tests (checking identical solutions) :)

[akshay326]

using QPs from http://qplib.zib.de/ Open to any other references!

[odow]

See https://github.com/jump-dev/DiffOpt.jl/issues/233 for a report and code that DiffOpt is much faster than alternatives. I think we need to double check what we're benchmarking and how.