Nonlinear Programming Benchmark Using COPS

Hans Mittelmann
Department of Mathematics
Arizona State University
mittelmann@asu.edu

Armin Pruessner
GAMS Development Corporation
Washington, DC
apruessner@gams.com

Acknowledgements

The authors would like to thank Prakhar Lodha, who wrote the parsing routines for the AMPL solvers. His routines enabled the results from the AMPL solvers to be used with the PAVER server.

Description:

This page describes the nonlinear programming benchmark results presented in the paper "A Server for Automated Performance Analysis and Benchmarking of Optimization Software". The performance analysis results were obtained using the PAVER server.

We used the GAMS solvers CONOPT and SNOPT, as well as the AMPL solvers IPOPT and KNITRO. The AMPL solvers were run through the GAMS/AMPL interface.

Our model selection came from an instance of the COPS test set. In particular, we used the models available in GlobalLib.

  • Download the paper: paver.pdf (1.5Mb)
  • See the PAVER results: all
  • See the PAVER results: subset of models
  • A list of the COPS models used. All models come from GlobalLib
    • . You can also download all models used: copsmodels.zip (4.4Mb).

    Reproduce the performance results using the PAVER server:

  • Download the data files: tracedata.zip (19Kb)
    • and unzip. You should have 8 trace files. The first 4 are for the complete COPS model test set The second set contains a subset of models showing the subjectivity of the model selection:
  • Submit the first 4 trace files and the last 4 trace files separately to the PAVER Server
  • You will receive your results via e-mail. The results should be similar to these results (all models,   subset of models) in the paper.

    • Last modified August 12, 2003.