Decide whether to include elitism

[NicMcPhee]

It turns out that ECJ doesn't include elitism by default (see a comment in simple.params). Given that elitism is widely used, and given that elitism could help narrow the gap between binary and size 7 tournaments, it seems that we might want to at least take a poke at it.

I think I'd like to try it on the K-Landscapes problem since there's considerable room for improvement in our performance on that problem, but we are getting somewhere.

[NicMcPhee]

I tried it on the Ordertree problem (which is so far turning out to be the most illuminating problem I've run things on), and adding elitism only made a very small (positive) difference on that problem. I'll probably try it on a couple of other problems, but if it doesn't seem to make much of a difference, I suggest we drop it, both to reduce the number of runs and to cut down on the complications of analysis and presentation.

[NicMcPhee]

The current runs are using elitism of both 0% (no elitism) and 1% (carry over the top 1% of the population). That does seem to make a difference, so we'll keep those two options in the pot moving forward.

[NicMcPhee]

In talking with Bill Langdon, he helped me realize that 10% elitism is really very big, and much bigger than most people would use. I'm guessing that such high elitism probably makes it quite likely that both parents are coming from the elite set, and therefore have similar-ish fitnesses, thus making XO bias much less effective.

Given the time, we don't want to re-run everything with a more "sensible" elitism, but we might re-run a few of them.

[NicMcPhee]

I've started a pile of Pagie-1/basic4 runs with 10K populations with elite set sizes of 10, so roughly one in a thousand instead of one in a hundred like before. We'll see what those look like and then I might do the same for something like the US Change problem.

[NicMcPhee]

@MKDramdahl suggested doing the Sine problem since that's where we started, which isn't a bad idea. I'll try to get that set up.