Parallelized hybrid optimization methods for nonsmooth problems using NOMAD and linesearch

G. Liuzzi, K. Truemper


We provide the following sections: It is also possible to download everything as a single zip file (beware that size of the zip file is approximately 81MB)

Convention of file names:

There are two types of file names for graphs.

  1. For the comparison of NOMAD, DENCON, and DENPAR as stand-alone methods, the following example serves as a guide.

    nomad.alone.dencon.alone.denpar.alone.noproc.50K.n500_30.pdf

    "noproc": run on single processor ; for 64 processors runs, this is replaced by "64proc" . "50K": limit on number of evaluations is 50,000 . "n500": NOMAD is run for 500 evaluations to obtain a starting solution for all three cases . "_30": problem set with 30 variables is solved. If the "_xx" part is missing, the entire problem set is solved.

  2. For the comparison of hybrid methods, the following example serves as a guide.

    nomad.nomad.nomad.dencon.nomad.denpar.noproc.50K.3K.n500_30.pdf

    "noproc": run on single processor; for 64 processors runs, this is replaced by "64proc" . "50K": limit on number of evaluations is 50,000 . "3K": limit on number of evaluations for each iterative application of each method is 3,000. "n500": NOMAD is run for 500 evaluations to obtain a starting solution for all three cases . "_30": problem set with 30 variables is solved. If the "_xx" part is missing, the entire problem set is solved.

Graphs for the comparison of NOMAD, DENCON, and DENPAR as stand-alone methods

back to top

Graphs for the comparison of hybrid methods

back to top

Graphs for the comparison of NOMAD.DENCON with 500, 1000, 2000, 3000, 4000 and 5000 evals/iter:

back to top

Graphs for the comparison of NOMAD.DENPAR with 500, 1000, 2000, 3000, 4000 and 5000 evals/iter:

back to top

Evaluation tools (as Matlab script files)

To reproduce the graphs listed in the Graphs section, simply run the script "do_the_plots" (from within a Matlab environment) and answer the questions that will be prompted.


Solver runs data files (as Matlab workspace variables):

back to top

Solver runs data files (as zipped text files):

back to top