Posted by David on July 27, 2003 at 06:49:16:
In Reply to: Re: mutations are too big posted by phil on July 27, 2003 at 04:30:48:
Well, first of all I modified the software recently so that now there is only a single crossover point with each mating. (The multiple crossover point version turned out to be buggy). So all possible combination ARE produced with each mating.
The bigger "problem" is that the crossover point is rarely the same with both parents. I did this to allow poem length to change, but it also leads to a fair amount of chaos; i.e. good genes are unlikely to be preserved. It has also led to this phenomenon of selecting for the shortest meaningful phrase. (However, I suspect and am starting to see evidence that these phrases will now begin to combine into larger meaningful poems.)
I'm definitely exploring evolutionary models for future versions and welcome all input.
--David
: it might be interesting if every possible child is produced in each mating. so 5 snip points are chosen, there would be 6 segments of each poem, and thus 2^6 children, each of which has the same chance of getting a mutation.
: : 2 parents should have 32 or 64 children. Each of this children would have only 1-3 random mutations.
: : types of mutation:
: : - replacing word
: : - new line insertion
: : - removing/adding word
: : People would vote for childrens to live.. there should only be 1-2 children left in the end.
: : This would speed up evolution substantially. In the matter of facts this is how things work in reallity..