October 31th, 1999: Joe is still alive, me to! Joe sent another two strategies, where Thornton seems to be semi good, but Hargrove seems to be real tough and such a good starter. I finally made another idea of mine ready: Curry . This is somehow halfway between NFocusPQ and Mimikri (or CopyCat, how they all name). You can either way. At least it is so much thrown out of NFocusPQ that I was very pleased by its success. It is also a rather friendly cooperative strategy. When I tryed to test variations of Curry against each other in the champions environment, they usually get to the top ranks, though only playing the same number maybe 50% of the turns! Find all three Strategies in the champions league !
September 15th, 1999: Two new strategies, Focus4 coming back. Well, it's gotten quiet arround numfield. Joes last strategy Assassin is a bit old, but he announced to fix one bug in my extensions of it. My latest research product is the Schizo family. Schizo actually is a meta-strategy to combine basic strategies. Both above strategies have for some reasons not been introduced in the regular tournaments yet. Focus4 comes back improved, after I removed an over adaption to some old test environment, which reverted its behaviour from a smart learner to a dumb Shark for a certain range of turns...
Anyway I decided to split the Numfield populations into two leagues . The champions league only contains the very best strategies. Of course due to the environmental effect, putting the best strategies into a separate pool watch them and see what happens. Pools also speed the simulation speed up considerably. Also it might be more appealing to newcomers if they could introduce their babies in less hostile environment. Hey, aren't there any newcomers? Well, why Joes and my strategies look so good. Because we both spent lots of time working on them. I guess one had do be fairly good to beat our top strategies, but look at our early experiments, do your early experiments compete with them? I guess the do!
Well, why are environments are so hostile anyway? Numfield was intended as a cooperation experiment. But, what happened? The richer set of possibilities (in comparison with the prisoners dilemma) and the interaction between all players at once, seems to make lots of learning necessary to adjust opponents behaviour. And when I manage to learn to predict the opponent. Why use it for cooperation, when I can use it to beat him? Thats what seems to have happened...
August 16th, 1999: Applet slows down. Retirements. Due to the number of strategies, their hunger for memory the applet had gotten somewhat slow, though still running on reliably JVM's. Well, to keep the applet running on smaller systems Joe offered to retire Ip, Alabama, Rosebud, SomeOne, Inigo, Vizzini, PlayItsafe. I gladly accepted. You can still see all of his strategies at Joes Page .
August 15th, 1999: Progress speeds up. I guess it is getting hard for any newcomer to come in this duell between Joe and me. Don't give up surviving for only a few hundret turns is probably better than Joes and my first dozend strategies achieved. Joe sent: Freberg, Lehrer, Ip, Yankovic, Alabama. Ip is meant to be a loosing variation of Pi and Alabama mainly aims at the Pi test. I introduce TimeWarp, this is also doing well in the Pi test, but isn't really finetuned yet. (Hmm, OK I tried, but I didn't improve it). I also introduce RandomPi, and WiseMan, a spontaneus idea for a friendly but opportunistic strategy. I withdrew Mimikri of mine.
Long term observation show that the success of the latest strategy is considerably blocked by the presence of those highly randomized strategies like Random11 and PseudoRandom12. Seems that those learning ones have problems adapting to random behaviour.
August 8th, 1999: The Focus4 Era is Over! Joe sent another update of Opus, which now is supposed to take much less memory, he also sent 4 new strategies: Vizzini, Rosebud, Inigo and Buttercup. Though sent as just two more strategies by Joe, the latter are very, very strong! Joe also withdrew LeadWatch. To honour Joe I have set up a page devoted to Joes strategies . To stay in business I have unleashed NFocusPQ which was supposed to be intermediate, but the meant to successor still performs well in the Pi test but not in the contest setting. Note the if Focus4 still performs well, this is now also due to it's similarity to NFocusPQ which establishes a weak form of cooperation between those two. Otherwise Focus4 gets beaten into ground by the Buttercup, Inigo, Rosebud, Storm gang.
August 2th, 1999: Well, I had reports, that this pages Applet is not running any more and therefore kicked out Const99, Const23, Const13, Tyranno5, Shark0 and Shark1. Potentially next candidates for deletion are Tyranno5b and Gentle Giant.
June 30th, 1999: Two strategies by Don Reble. Pi was not meant to be serious, and you can see at the Pi-Experiment page , why this strategy is interesting anyway! Three new strategies and one update to Ohio by Joe Huber. Announcement: This environment is getting a bit large, there are many boring stratgies, there is also Joe's new Opus, which uses a lot of memory (quadratic in th enumber of participants). A predecessor of Opus was named Elephant according to Joe and it just blew my Java Virtual Machine... So what I will do somewhere soon is to withdraw the following of my strategies: Const99, Const23, Const13, Tyranno5. This will affect the outcome of the first rounds and therefore change simulations quite a bit. Be prepared.
BTW in case there will me more contestants somewhere in the future might had to limit the number of introduced strategies per player to s.th. like 2 or three per pool. One reason for this is the environmental effect demonstrated in more detail on the Pi Experiment Page .
If you have any other question, wishes or ideas, strategies or even class files please send an email to dorisnfrank@bigfoot.com .