Flower/WinThreshold

I find it strange that Go-Players shun draws to such an degree that special care is taken of eliminating the slightest possibility of drawing by introducing fractional komi and adjudicating 0.5moku advantages as a 'win'. Aside from thinking that any rating system would benefit from assessing these situations as a draw, I go even further and prefer that a certain threshold needs to be exeeded in order to consider the game a win. (For example 4Zi/8Moku) Thus small differences in score no longer decide between win/loss but rather between win/draw. Draws (which are win/win) occurr more often and I am happy =]

Note: The value of the threshold would depend on the skill of the involved Players. Initially I intended to include some elaborations on rank dependend scaling of the threshold, but for simplicities sake as well as the limited scope of its application I have decided to give an example by noting the threshold I use at the time of writing (I am KGS22k). I view the threshold as a tool to mask the noise floor in one's skill. The correct level depends on the amount of noise that is present. At Beginner Level the threshold would be large and reduce as the 'instability in skill' or 'noise' of ones moves reduces. As the discussion thread says: if Pro's were to use a threshold of 8moku then it would not only mask the noisefloor but also elegant strategy and tactics. In order to win a Pro would need to resort on more 'coarse' strategies. In case of Professional Dans it is pretty safe to assume that the appropriate threshold would be near zero moku.

last updated 2007-01-18

/discussion?

Phelan: I think the issue with avoiding draws by a half-point komi is mostly because of tournaments, where it would make both selecting opponents for next rounds, and the final round more complicated.

Flower: Ahh yes This might be true. There would be a clash of interests in an accurate rating system that tries to space contestants according to their relative skills and a tournament system which tries to determine a strict hierarchy of finalists by elimination.--Flower, 2007-01-17

Herman Hiddema: Making games with a difference <= 8 draws would make a very large percentage of pro games draws. Take for example te 11th LG cup ( http://gobase.org/games/nn/lg/11/ ). Of the 28 games finished so far, 13 were won by resignation. Of the remaining 15 games, only 2 were won by more than 8 points. Even assuming that all resigned games would have ended with more than 8 points that would mean that almost 50% of these games would have ended in a draw. In my opinion, the rarity of draws even at high levels is one of the advantages of go over chess and draughts/checkers. Furthermore, the correct komi is not actually known.

Blake: Even if you just look at high-level players on KGS, the same thing holds true. Seonga, for example, plays at 7-dan. In January, most of his games have been resignation wins or losses; however, his wins on points have 7.5, 0.5, 5.5, and 11.5. abel, a six-dan, has several half-point wins and single-digit wins. As you get stronger, the accuracy of your counting increases, and the significance of each point grows; calling a single-digit win insignificant isn't really sensible in this context. Now, as for half-point wins, that's just a way of saying "white wins draws." It's a method of balancing the fact that Black moves first.

Alex: The main advantage of introducing something like this at pro level is that you'd get longer, more tightly fought games; a player with a five-point lead couldn't sit comfortably on it in the late middlegame, but would have to press his advantage and try to rack up enough points for a win, while comfortable in the knowledge that even a ten-point blunder would still result in a draw. Meanwhile, a player losing by 10 would not have to start looking for last-ditch gamble before resigning, but merely ramp up his aggression slightly and hope to bully his way back into draw territory.

The downside is that it would result in a huge bias towards fighting players, since their wins are often by resignation, whereas the more calculating players wouldn't be able to play their "keep it close until endgame and eke out a small win by superior counting" strategy, because all their wins would be draws instead. Given that the world's strongest player, Lee Chang-ho is one of this type, I think that would be a huge tragedy. I've been studying his game against Yamashita Keigo in the 19th Fujitsu Cup, and he makes a brilliant probe on move 20 that soon results in him taking the lead, and by move 58, he's in full "lockdown" mode, counting the score and playing as conservatively as possible to preserve his lead, resulting in a 2.5 point win which, the commentators say, was never in question. Despite the small margin, Lee never gave Yamashita the slightest window of opportunity to get back in the game. Some might call such a game "boring," but I think it's brilliant. It would be a shame to force everyone to fight all the time.

Anyway, I think there's a certain elegance to the notion that all you need is "half the board plus one" to win. Even komi breaks that simplicity somewhat, but it's arguably a necessary evil. I don't think the merits of this idea are worth it - "if it ain't broke, don't fix it," right?

Flower: Thanks for your elaborate pro and contra Alex!:-) I would agree that while the threshold is an interesting idea it would change (even if scaled properly) the game strategy and is thus not fit to be implemented on broad scale without further reasons. Yet what led me to impementing it was far more critical ;-) My partner was crestfallen due to a series of several near losses and had stopped our weekly WeiQi sessions. The time was grim indeed and he might have ceased to play WeiQi forever if I had not suggested that we consider close results a draw. Surely you agree that it is of paramount importance that I safe a soul from falling back into the darkness? (as well as my poor Goban from gathering dust) ^-^ --2007-01-18

Velobici: Ratings are based upon The Method of Paired Comparisons. Draws are rather difficult to handle in that statistical framework. This is a reason that draws are shunned in any rating system. Elo was forced to distort/modify the method in order to deal with the numerous draws in chess. More information available at:

• Book Citation: http://spartan.ac.brocku.ca/~lward/Thurstone/Thurstone_1927a.html
• Citation: http://links.jstor.org/sici?sici=0040-1706(198911)31%3A4%3C495%3ATMOPC%3E2.0.CO%3B2-Z
• Journal: http://biomet.oxfordjournals.org/cgi/reprint/77/2/265.pdf
• Journal: http://biomet.oxfordjournals.org/cgi/content/abstract/74/2/432
• Short Article: http://eom.springer.de/p/p110050.htm and
• Short Description: http://www.fs.fed.us/rm/value/The%20Method%20of%20Paired%20Comparisons.html

The first link, H. A. David's book is the work to start with to understand ratings systems.

Flower: Thanks for the links Velobici! Though I am not completely clueless about Rating systems, I would not presume myself to be fit enough to discuss them on academic level. My experience with them stems largely from migrating Elo Systems to Bayesian Systems based on Glickman's research. I never encountered severe problems with draws though. Regarding Elo I found his method of treating a draw as 0.5 result very elegant and until now I have not encountered problems when comparing the result to the exspected outcome to calculate the Elo-rating adjustement. Yet lacking true in-depth understanding I could not infer that draws are truly unproblematic on the fundamental statistical level. --2007-01-18

Flower: Thank you all for so many replies! :D / I certainly had not in mind proposing it to the Pro world or even the broader public ;-) For a lowly beginner Kyu as myself (and his games with his friends at home) it helps to reduce these 'random wins'. Let us not forget that there are many different levels of playing Go. Beginner Play and Professional Play differ in such a great degree that I am amazed that my suggestion was assumed to be for Pro-play. I think it would be a good if rulesets were designed in a more modular way including target audience specific scaling of certain variables(e.g. komi). I guess Ikeda's Rules and the Mailing List's International Rules are a good start ^-^ --2007-01-18