(Negative) Impact of root XO points - NicMcPhee/MICS-2014-GraphDB-EC GitHub Wiki

It occurred to me that it wouldn't be hard to write a query that would let us find out if there were any instances of a root XO or mutation point (i.e., replacing the entire "root" parent with a subtree from the other parent, or a randomly generated subtree). Unfortunately running the entire query seemed to be taking forever, so I limited it to individuals in the last 11 generations:

match (n {generation: 100})<-[:ELITISM|PARENTOF|MUTANTOF|ROOT_XOOF*0..10]-(p {cutPoint: 0})
return count(distinct p);

I ran the query on the 10K individual run, and found that there were only 87 root cut points that had root-path descendants in the final generation, and they were all at or very closet to the final generation, and of those 3 were from generation 95, two from generation 97, two from 98, and 4 from 99; all the rest were from generation 100. Only one of these had a penalized fitness less than 9, and many were well above 100, so it's clear that this crowd wasn't going to be the dominant force in the evolutionary process. This makes it quite clear that (at least in this run) root cut points were definitely not successful, at least in terms of having offspring in the final generation.

Given that the fitnesses of those few individuals from the previous query weren't very impressive, I thought I'd see pull the median penalized fitnesses of root cut individuals that were root-path ancestors of individuals from the final generation:

match (n {generation: 100})<-[:ELITISM|PARENTOF|MUTANTOF|ROOT_XOOF*0..10]-(p {cutPoint: 0})
return percentileCont(p.penalizedFitness, 0.5);

The result was a pretty Not Good value of 195.32. I did similar queries for cut points of 1, 2, 3, and 4 and got values of 48.46, 40.99, 15.46 and 32.75, all of which were substantially better. Note also that the results for cut points of 3 and 4 were substantially better than for 1 and 2, which suggests a (not surprising) trend for less damage as the cut point moves down the tree, but not uniformly as 3 had much better fitnesses than 4 did.