Optimization using Biomorpher

Replies to This Discussion

Permalink Reply by John Harding on February 20, 2018 at 5:25

Hi Abdulaziz,

Pleased to hear you are using biomorpher. To summarise the current process:

At each generation, fitness is assigned to each phenotype from 0.0 to 1.0. These are then used for roulette wheel selection (see Goldberg's classic book for more info on this). There is no elitsm at all. The process is as follows:

• All fitnesses are reset to zero.
• If a manual selection is made (i.e. the tickbox checked), than all designs in that cluster will have fitness set to 1.0 regardless of performance-based fitness. They can go no higher.
• If no performance-based criteria is specified, goto selection.
• If one performance-based criteria is specified, then normalise performance values for the whole population and assign this as the fitness. If fitness is already 1.0 (due to manual selection) then do nothing. Note that if you are minimising a performance value, we normalise and take 1-x of course.
• If two or more performace-based criteria are used, then simply normalise performance values, sum these and divide by the number of criteria.

So at present, there is no Pareto approach here. Multiple criteria are assumed an equal weighting based on the min and max performance values, and combined to a single objective. Academically, this is known as a 'no-preference method'. This will likely change in future as we develop the tool, but even mixed mode interactive and performance based optimisation is quite rare, so we'll see how things go and how much time I can find!

Just to add, this is different to typical multi-objective optimisation methods in that you could in theory evolve something that is poorly performing for all performance-based objectives (i.e. nowhere near a Pareto front), simply because you like the look of it.

I'm just writing a paper on this at the moment, so will have some more references in due course.

Hope this helps,

John.

Permalink Reply by Abdulaziz Afandi on February 25, 2018 at 8:55

Hi John,

First, Thank you for your answer this is very help full. I would like to read your paper once you publish it.

Looking to hear from you soon,

Abdulaziz