Metaheuristics for Clojure

Metaheuristics for Clojure is a Clojure library for common (and some uncommon) metaheuristic algorithms. It uses Clojure Agents to parallel compute the fitness functions, so there will be a distinct speedup on multi-core machines. It was tested on a dual 4-core Intel Xeon (W5580), where it ran 6.3 times faster than on a single core machine.

These algorithms are implemented at the moment:

• Particle Swarm Optimization (PSO), including some extensions (velocity clamping, inertia weights, multi-start resetting)
• Genetic Algorithm (GA)
• Evolution Strategies (ES)
• Invasive Weed Optimization (IWO)

For more information about these algorithms and techniques, I recommend these sources:

• Introduction to Evolutionary Computing - Eiben/Smith
• Fundamentals of Computational Swarm Intelligence - Engelbrecht
• A novel numerical optimization algorithm inspired from weed colonization - Mehrabiana et al. (paper)
• or the wondeful, free book from Sean Luke: Essentials of Metaheuristics

Usage

The algorithms live in the namespace 'metaheuristics.*', where * stands for one of the abbreviations from above. There is only one public function to call for each variant (same abbrv.). They all return a struct with the global best solution. The fitness value and the position ca be accessed by (:position gbest) (it's (:chromosome gbest) for the GA/ES) and (:fitness gbest).

The GA und ES return a genotype representation (think of it as an internal representation of the solution). You have to convert it to a phenotype representation with the (chromo-to-phenotype) function. E.g:

(chromo-to-phenotype (:chromosome gbest))

See the code for more information. It's not that complicated.

Known issues and limitations

For now, the functions can only handle real value solutions. Binary, decimal and permutations will be added later.

Also, I am by no means a Clojure expert. Bugs, bad design...it's possible that you will find all of that in my code. If not, phew!! :)

Installation

• If you want to work with the source: It's a Leiningen project. From here on I think you know what to do.

• If you just want to use the library in your own project: Use the Leiningen (see above) to compile a .jar and put it to the path of your own project.

License

Creative Commons - BY-NC-SA

Feel free to contact me for any questions.