New Solver (LP, QP, MILP, NLP, and MINLP) for Pyomo

[John Hedengren]

Dear Pyomo Community:

Attached is a Python solver interface to the APOPT solver. Our group at BYU (thanks to Logan Beal) wrote the interface in Python so that it would be compatible with MacOS, Linux, and Windows. It is free to use for commercial or academic use and can be installed by placing the apopt.py file in your path and selecting apopt.py as the solver. We are still testing the MINLP functions but LP, QP, and NLP features should work fine. It sends the NL (model) file to byu.apopt.com and returns a solution (sol) file. There is no need to compile the solver - it works with either a Python 2.7 or 3+ distribution as a web-service.

Additional information:

https://en.wikipedia.org/wiki/APOPT

Git Repository

https://github.com/APMonitor/apopt

I’m interested in feedback on the performance. This summer we are also working on combining the active set SQP (APOPT) and interior point (BPOPT) solver methods. Some preliminary benchmark results are shown below on a set of ~500 benchmark problems (APOPT+BPOPT).

I'll send out another update when the MINLP features are working.

Best regards,

John Hedengren

[Bill Hart]

John:

Thanks for the contribution!  We're considering a move to github soon.  I'll integrate your solver interface when we finalize that plan.

--Bill