optproblems 0.2

Infrastructure to define optimization problems and some test problems for black-box optimization

Description

This package contains a collection of common benchmark problems for black-box optimization. Under the term “black-box problem”, we understand problems for which only little is known about their structure and properties. Such problems usually appear in practice when simulator output or some other complex system with nonlinear behavior is to be optimized.

Contained test problems:

• Binary problems OneMax, LeadingOnes, and LeadingOnesTrailingZeros

• Multiple Peaks Model 2

• Walking Fish Group (WFG) toolkit

• ZDT collection for multiobjective optimization

More test problems shall follow in the future.

The infrastructure of this package can also be used to wrap your own (real-world) optimization problems in the problem base class. Reasons to do this may be the following features:

• Support for single-objective and multi-objective problems

• In general, no assumptions about the search space are made

• Evaluations are automatically counted

• Can use true parallelism or concurrency via multiprocessing(.dummy)

• Provides functionality for checking bound constraints and repairing violations of them in continuous optimization

• Optionally: detection and subsequently avoidance of duplicate evaluations

Dependencies

Some modules of this package have additional dependencies on third-party packages. However, these are not enforced during installation not to hinder the utilization of the base module, which will always be free of dependencies.

Module	Dependencies
base	none
binary	none
mpm	numpy
multiobjective	diversipy
wfg	diversipy
zdt	diversipy

Documentation

The documentation is located at https://ls11-www.cs.tu-dortmund.de/people/swessing/optproblems/doc/