MOPEDS

Multi-Objective Problem and Experimentation Data-Structures

Disclaimer

This is my personal research repository for benchmarking multi-objective algorithms.

Please don't consider anything stable you see here. Interfaces might change, and parts of this package could be split out.

Goals

1. I hope to develop a common interface for many (nonlinear) multi-objective test problems.
2. Curate a library of test problems.
   Problems can be unconstrained, have variable constraints, linear inequality or equality constraints, or nonlinear inequality or equality constraints.
3. Facilitate interoperability between different solvers and different packages.
4. Provide ways to compute common performance indicators.

Current Status

I have spend way to much time thinking about 1) + 3), so there are no problems here, yet. But I think, the bridging mechanism is really really neat!

Related

Ours

• NonlinearCG package: A repository with some nonlinear CG algorithms and MOO test problems.
• This repository has code to wrap the Fortran test problem library TESTMO.
  We compared Compromise.jl against DFMO.
  There were a lot of awkward code manipulations. This wrapper is much nicer!

Others

• Metaheuristics.jl has algorithms, some problems, and performance indicators!
• pymoo is a Python package with test problems and algorithms.
• MathOptInterface.jl provided most of the important ideas concerning bridging.
• DifferentiationInterface.jl inspired and influenced the caching mechanism. We support automatic differentiation.
• MultiObjectiveAlgorithms.jl, a collection of MOI-backed algorithms.

TODO

• Obviously: Fulfill goals stated above.
• Documentation and examples.
• Adapt predecessor code in ./src/NonlinearMultiObjectiveStructures.