Sub-page of FairKomi

Herman Hiddema: What makes you say that correct komi is zero for random play? In random play, black still starts and has half a stone more on the board on average, which is an advantage...

Bill: Yes. Correct komi for random play is probably a little less than 1/2 correct komi for perfect play. (Usually the worst loss from random play is small. OC, it can be spectacularly large.)

Flower: I read it somewhere on Sensei's that komi for random play is near zero. But thanks for contesting this. I just assumed it to be true. I think I remember the one who wrote it that it can be verified by letting computer programs making random moves play against each other a large number of times :-) I assumed it to be likely as to my mind the first stone will only tilt the winning chance in favor of black if black has the ability to defend this advantage. (which would be difficult given random play)

LukeNine45: Actually, I bet one could write a program that would play quite a bit worse than just random, one that intelligently TRIES to lose. What would proper komi be if two of these players play? I guess if both pass right away it'd be a tie and komi should be 0 to make it fair? Hmmmmm...

Flower: Hehe :-) Yes I can see what you are getting at. And I agree that my analogy of 'worst possible play' to 'random play' is void. I do not exactly know with what to replace it though: Hmm perhaps 'Worst possible play with the aim to win?'. I Guess I will remove the 'worst possible' part for now.

blubb: In my view, it makes sense to treat random as the natural "zero" of the rank spectrum. The commonly used ranks are probably off by something between 120k and 180k. (Because of the nonlinear effect of handicap, about 230 to 350 no handicap, no komi tournaments would be necessary to evaluate this offset reliably.)
Play below zero behaves quite strangely. Supposedly, the worst possible play (under area or stone scoring) that doesn't require knowledge about the opponent is "always pass", which then constitutes the opposite of perfect play.

MrTenuki: Actually, "always pass" is not quite bad enough:

Even worse than simply passing  

blubb: First off, I wrote "under area or stone scoring" for a reason. :) Under neither this will be worse for Black than passing.

Secondly, even under territory or prisoner scoring, this kind of play does assume something about the opponent's play, like, that White goes ahead to capture those stones. An otherwise perfect player with additional personalized knowledge generally has an advantage over a mere perfect (but opponent-neutral) player and may yield better results - in Anti-Go just as in normal Go. If here, for example, White did nothing but pass, one could say that black still would stand better (at the very least, not worse).

(English grammar question by a non-native speaker: What's the correct x tense in "If A did something, one could say that B x something else."?)

LukeNine45: I believe you're looking for "B would do something else." And the point you're making is the reason why I force both players to move once in my Go variation. Things get very odd when both players are trying to lose...

blubb: Ah, so the conditional isn't closed by "could say" but embraces x - thanks. About Gopposite, I have replied there.

LukeNine45: You've inspired me: Gopposite. hehe.

Flower: Would anyone know a program that allows easy computer tournaments between Go Programs? (they should habe no problem with utilization of the Baka no Itte principle :) I am looking for something that could play large number of games unattended and present me with a outcome list (including the mean result if possible)

blubb: I have concerned myself with estimating (1) the rank of random and (2) proper komi over the rank spectrum two years ago, and the data indeed seemed to indicate that komi has to be smaller at lower ranks. I couldn't afford the computing capacity really needed to finish the project though. If anyone with access to more than a simple PC was interested, I'd like to get it done, of course.

Bill: Well, anyway, if human novices know enough not to commit obvious harakiri (as in the next diagram) already the komi should be at least half the komi for perfect play, I think.


From my experience, komi becomes meaningless around 18k because games are now rarely close. -JH

FairKomi/discussion last edited by HermanHiddema on April 14, 2009 - 16:39
RecentChanges · StartingPoints · About
Edit page ·Search · Related · Page info · Latest diff
[Welcome to Sensei's Library!]
Search position
Page history
Latest page diff
Partner sites:
Go Teaching Ladder
Login / Prefs
Sensei's Library