Diese Seite gibt es leider noch nicht auf Deutsch. Ich bin dafür, dass sich das ändert!

Numfield, the schizophrenic Schizo

Schizophreny --- having ones mind divided into several independent persons.

Well, the name is not entirely well chosen. The basic idea of Schizo was, that when you take bunch of a few (imperfect) strategies watch their behaviour and learn which ones to trust and let them vote for next turn, one must achieve an improvement. But this is only an heuristic idea. There is no proof or what that such a kind of adaption idea might work: A poor adaption might select the poorest of all bids at every turn and therefore be worse than any of the single strategies.

Schizo is an implementation of that idea. And it looks like the choosen selection scheme works reasonable. For simplicity Schizo uses instances of the other strategies as plug ins. Therefore it needs more memory and time than any of its partial strategies.

Schizo comes with a bunch of close Relatives:

• Schizo itself is based on the 6 top strategies NFocusPQ, TimeWarp, Focus4, Freberg, Lehrer and Opus.
• SchizoF is based on Franks 3 top strategies NFocusPQ, TimeWarp and Focus4.
• SchizoJ is based on Joe Hubers 3 top strategies Freberg, Lehrer and Opus.
• SchizoL is based on less successfull strategies, to test how much adaption can gain from those: Random11, WinningMode, Pi, Ip, GentleGiant, Shark7, Nihilist.

B elow you can see how successfull the variations on Schizo are in an Champions League environment. Find the source code of Schizo on the Source Code Page . I see now reason to introduce Schizo in the regular tournament, since it is no real success, no real new algorithm, just a blend. You need Java to see s.th. in this area!

I lend it to you to make any conclusions from the above. I have to say I didn't expect what eventually happened... It seems like different strategies gain differently from the adaption.

If you have any other question, wishes or ideas, strategies or even class files please send an email to info@doris-frank.de .