After close examination of the tactics employed by the best individuals from successive generations, I noticed that strategies that were the best in early generations got superseded, as you would expect, but then reappeared several generations later!
While this at first was somewhat puzzling there turned out to be a logical explanation: If program x is always beaten by program y, and y by program z, then that does not mean that x cannot beat z. In practice it was more like this:
- program a from population A is randomly selected
- program b from population B beats a
- c from A beats b
- d from B beats c
- e from A beats d
- f from B beats e
- g from A beats f
- ...
- m from A beats l
- n, equivalent to f, from B beats m
- repeats from step 7.
So this was why the simulation never produced strong players.
Next, I will describe how I solved this problem, but I'd be interested to hear from you if you have any suggestions.
No comments:
Post a Comment