engine inquiry

Replies to This Discussion

Permalink Reply by John Harding on September 16, 2017 at 8:52

Dear Anas,

To answer your questions in sequence:

• The genetic algorithm uses a floating point encoding, with single point mutation. There is only one probability rate which can be triggered on any gene. Mutation at present simply replaces the gene with a new random (normalised) value. In future I should implement polynomial mutation or similar, but to be honest I have been trying to keep things as simple as possible. Selection is roulette wheel for both artificial and performance based methods.
• There is no crossover at all. Indeed, crossover has been shown to be somewhat ineffective for interactive evolution (in general cases), although I don't have the reference to hand (note: If you use Biomorpher with Embryo, things are a little different as cross-over in the definition can be caused by a single point mutation in the genotype, but let's not go there here).
• Clustering algorithm is K-means++. The cluster centroid is actually the closest member of the population to the true centroid. We may change this in future, so that the centroid itself can be a new design, but not yet. Google 'Cluster-Orientated Genetic Algorithms' for more info, and a couple of papers written in 2004 I believe.
• No publication as yet. We are planning something though after a couple of upcoming workshops. It will be based on mixed-mode evolution (objective function with artificial selection in the same run). Will post something when I get round to writing something!

Nice to hear you are using biomorpher, what do you plan to use it for in the end?

Thanks,

John.