[ANN] Watchmaker Framework for Evolutionary Algorithms version 0.3.0 released

K

Kent Paul Dolan

Daniel said:
Daniel Dyer
https://watchmaker.dev.java.net - Evolutionary Algorithm Framework for Java

Well, it works, but without crossover, the TSP example
is very weak. Without elitism it doesn't work very often
at all.

Do you have descriptions somewhere for your list of
selection mechanisms? There are a couple I don't
recognize.

xanthian.
 
D

Daniel Dyer

Well, it works, but without crossover, the TSP example
is very weak. Without elitism it doesn't work very often
at all.

The TSP applet is not intended to be anywhere near as comprehensive as
your own work on the problem. For example, my brute force implementation
is very naive. It does serve two purposes though. Firstly as a
demonstration of how to use the components in the framework to solve a
problem and secondly as a demonstration of the effectiveness of
evolutionary appoaches. You are probably right though, with cross-over it
could be an even more effective demonstration of the power of evolution.
It was something that I initially intended to include but, for whatever
reason, I didn't. I may well add this in the next version and see what
improvement I get.

Did you implement ordered cross-over as defined on this page
(http://www.permutationcity.co.uk/projects/mutants/tsp.html)?
Do you have descriptions somewhere for your list of
selection mechanisms? There are a couple I don't
recognize.

I think they are all in the Mitchell book.

Stochastic Universal Sampling is a fitness-proportionate strategy similar
to Roulette Wheel Selection except that it ensures that observed selection
frequencies are in line with expected frequencies. It is most useful for
small populations where statistical anomalies could skew the results
considerably if Roulette Wheel Selection was used.

Rank Selection assigns selection probabilities as a function of the
relative fitness scores rather than absolute scores. This can increase
selection pressure when there is little variation in the population or
reign it in when the fitness scores vary wildly. My default mapping
function is very trivial (for a population of size n, the fittest
candidate gets a score of n, the next n-1, etc. and selection is
proportional to these scores). This default linear function can be
over-ridden with something else (maybe in some situations it would be
better for it to be exponential or logarithmic).

Dan.

(I have removed comp.lang.java.programmer from the follow-ups as this
isn't really on-topic)
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Members online

No members online now.

Forum statistics

Threads
473,743
Messages
2,569,478
Members
44,898
Latest member
BlairH7607

Latest Threads

Top