| Alex M |
Leo and I worked on the roulette algorithm to figure out why the new individuals being formed are different in the asymmetric bicone version of the loop than the original and database versions. We made some fixes, but we came across a complicated problem. Essentially, we think that the difference is coming down to the different number of times the random number generator is being called. Oddly, we can get the lengths that are generated to be the same. But getting all of the genes to be the same would require substantial reworking of the structure of the roulette algorithm and how to do it isn't obvious. We can keep trying to figure it out, but the restructuring was important to allow the mutations to occur on either chromosome (vs in the original and database versions, where we just have one chromosome to mutate, here we need to allow for more mutations since we have up to 6 genes). In principle though, I feel like this is possible, just difficult and requires substantial rewriting. |
I'll keep looking at the roulette algorithm to see how we can get the same new individuals as the previous two runs. Julie and I were also planning on meeting with Ben about AREA to better understand how to run it properly, |
| Ryan |
I fixed the bugs that I was having yesterday and the algorithm should run more properly. I also added functionality for sending certain parts per ten individuals to either tournament or roulette. However, neither function is getting "optimized" results by 100 gens (though the tournament is significantly closer than roulette in isolation) with the highest fitness score I have been able to get was ~8.65. But, I also do not know what the theoretical max fitness score, given the parameters, would be. |
Find out what an optimized fitness score's value would be so I know how close I am to achieving that. Run preliminary tests to see what combinations work best for now. |