This repertory contains the source code of the DMulti-MADS algorithm for constrained blackbox optimization. In terms of performance, it is more efficient than the old implementation found in [DMulti-MADS][https://github.com/bbopt/DMultiMadsEB].
Jean Bigeon, Sébastien Le Digabel and Ludovic Salomon, Handling of constraints in multiobjective blackbox optimization
Warning : This code has no vocation to be used in industry, see Nomad for a more robust implementation of state-of-the-art blackbox method. It aims at guaranteeing the reproducibility of the experiments described in this work.
To use DMulti-MADS, Julia >= 1.6 is required. One can test it by typing the following command at the root of the directory.
julia> ]
(@v1.6) pkg> activate .
(DMultiMadsPB) pkg> test
All tests should pass.
A simple example can be found in the examples/ folder. One can look also at ./test/madsmodel.jl for more examples.
This folder contains an implementation of all multiobjective benchmark problems used in this article for the Nomad (BiMADS) software.
The algorithm [DFMO][http://www.iasi.cnr.it/~liuzzi/DFL/] is provided with all benchmarks coded in Fortran by the authors. For this reason, it is not given here.
Warning The generation of analytical benchmarks takes a lot of time (around three days and requires more than 40 G of memory on hardware;
For BiMADS, all executables are given with models and nelder-mead search deactivated. Uncomment the lines in the main function if you need them.
To obtain the real blackbox optimization applications, one can get them at:
- [STYRENE][https://github.com/bbopt/styrene]
- [SOLAR][https://github.com/bbopt/solar] You can compile them and launch them on the different algorithms with the scripts provided (minus paths to adapt to your machines):
Warning Solving STYRENE and SOLAR for a given solver takes one day.