Multiobjective PSO or ACO available in pymoo?

[PBerit]

Hi all,

I would like to know if other types of multiobjective optimization algorithms like Multi-objective particle swarm optimization (MO-PSO) or multi-objective ant colony optimization (MO-ACO) are available in pymoo. On the list of the website I could not find them and I am wondering why they are not there. From what I read it definitely makes sense to also use MO-ACO and MO-PSO for testing on a problem instead of only multi-objective evolutionary algorithms as pymoo mainly uses.

[jacktang]

Hi @PBerit, currently pymoo only supports SO-PSO without constraints.

[PBerit]

Thanks jecktang for your answer. Do you know if there are any plans of including PSO or ACO for multiobjective problems (with constraints) in pymoo? So far not many state-of-the-art algorithms are supported by pymoo for multiobjective optimization.

[blankjul]

What MO-PSO versions are you referring to? SMPSO or OMOPSO?

ACO has so far not been the focus of pymoo.

[PBerit]

I am referring to multi-objective particle swarm optimization in general. Is there any included in Pymoo?

[blankjul]

Currently not. Are you a researcher in this area and would be interested in contributing? I would be interested in adding one or more versions of PSO if you are interested.

[PBerit]

Thanks for the answer blankjul. I am not really a reseacher in that field altough I have a basic understanding. I just apply these methods without knowing too much about the details. I know that other frameworks for multiobjective optimization in Python have implemented it (e.g. jMetalPy https://github.com/jMetal/jMetalPy). So I was wondering why Pymoo does not include multiobjective PSO as it is one of the main metaheuristics for multiobjective optimization. For me it is kind of strange that it is not included and a big disadvantage, compared to other frameworks. Are there any plans from the Pymoo-team to include it soon?

[blankjul]

Probably not soon, but it definitely is a good consideration. If someone creates a PR reimplementation of the jMetaPy version I am more than happy to look at it.

Have you benchmarked the version you are referring to? Is it outperforming the NSGA-II implementation in pymoo?

[PBerit]

@blankjul : Thanks for your answer blank. What do you mean by PR reimplementation?

Actually I have not benchmarked the multi-objective PSO. As you know, the quality of metaheuristics strongly depend on the problem. So for some problems the evolutionary algorithms might work better than particle-swarm optimizaiton and for other it's vice versa. I think you find some problems in the literature where particle swarm optimizaiton is better than evolutionary algorithms (and vice versa). But okay, I understand that you don't have time to implement it and you choose to only implement evolutionary algorithms. I'll see whether I proceed using Pymoo or switch to jMetalPy. Thanks for the information.