Obtaining Quadratic Models of Noisy Functions

Aswin Kannan and Stefan M. Wild

warning

This page contains supplemental images for the paper:

Obtaining Quadratic Models of Noisy Functions by A. Kannan and S. Wild, Mathematics and Computer Science Division, Argonne National Laboratory, Preprint ANL/MCS-P1975-1111, November 2011.

Read these general instructions for comparing the solvers. The entire package may be downloaded from noqs.tar. The solver 'minq' can be obtained from here. The solver 'qppal' may be obtained upon request to Dr. Sven Leyffer ([email protected]). The corresponding solver paths have to be changed in the script 'testsolver.m' (lines 76-78) in the 'data' folder.

Test data:

(MATLAB files)

Driver script to define problems matgen.m
Problem defining function getpoints.m
Test set m (BCQP assumptions apply) minqset.mat
Test set r (BCQP assumptions do not apply)
- Overdetermined problems overdet.mat
- Rank-deficient problems rankdef.mat
- Degenerate problems degen.mat
Other problems (scalability) scaldat.mat

The noqs solver:

Readme readme.txt

(MATLAB files)

Main function callnoqs.m
Pre-processing routine getphi.m
Linear feasibility solver linfeas.m
Active set solver noqs.m
Post-processing routine postprocz.m

Comparison results:

(MATLAB files)

Benchmarking script testsolver.m
Infeasibility checking routines infcheck.m and infsol.m and distance routine distnm.m
Test Set m - results (BCQP assumptions apply) minqsetcomparison.mat
Test Set r - results (BCQP assumptions do not apply)
- Overdetermined results overdetcomparison.mat
- Rank-deficient results rankdefcomparison.mat
- Degenerate results degencomparison.mat
Other results (scalability) extrascal.mat

Additional images:

(MATLAB files)

Image generating script noisyqpplots.m
Performance profile subroutine perf_profile.m