Comments on: Genetic Algorithms explained… http://www.genomicon.com/2009/01/genetic-algorithms-explained/ The Crowd-Sourcing of Intelligent-Design Wed, 08 Feb 2012 01:08:46 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: Mesh Wars : How to make Skynet http://www.genomicon.com/2009/01/genetic-algorithms-explained/comment-page-1/#comment-1005 Mesh Wars : How to make Skynet Tue, 26 Jan 2010 01:51:18 +0000 http://www.genomicon.com/?p=444#comment-1005 [...] Genetic algorithms are simple little programs that simulate evolution by (re)combining groups of actions, keeping the ones that works best, ditching the rest, then recombining again. They can arrive quite quickly at solutions to complex problems. [...] [...] Genetic algorithms are simple little programs that simulate evolution by (re)combining groups of actions, keeping the ones that works best, ditching the rest, then recombining again. They can arrive quite quickly at solutions to complex problems. [...]

]]> By: admin http://www.genomicon.com/2009/01/genetic-algorithms-explained/comment-page-1/#comment-12 admin Sun, 04 Jan 2009 00:28:44 +0000 http://www.genomicon.com/?p=444#comment-12 What this seems to do is get very close to the goal... ie: get a stable population that are of equal merit but still not quite right (and these are often the same "solution"), then wait until a mutation finally hits on the correct value... so the type of mutation is quite important. In the streamlined zip version above, I've changed the mutation from being just another random number to a random 1 or -1, which makes it far more efficient. I also assumed from the original example that what was required was an entire population of correct results rather than just a single correct result - either are valid, but the streamlined version stops when it gets to the first correct result. You can also tweak the efficiency by changing the mutation and population sizes. I think I might make one of these that tests the inputs of another one of these to see which is the optimal combination. What this seems to do is get very close to the goal… ie: get a stable population that are of equal merit but still not quite right (and these are often the same “solution”), then wait until a mutation finally hits on the correct value… so the type of mutation is quite important.

In the streamlined zip version above, I’ve changed the mutation from being just another random number to a random 1 or -1, which makes it far more efficient.

I also assumed from the original example that what was required was an entire population of correct results rather than just a single correct result – either are valid, but the streamlined version stops when it gets to the first correct result.

You can also tweak the efficiency by changing the mutation and population sizes. I think I might make one of these that tests the inputs of another one of these to see which is the optimal combination.

]]>