![[Welcome to Sensei's Library!]](images/stone-hello.png "Sensei's Library")
Moved by aceofspades from Komi mainpage on 30.12.05.
This is a different idea of mine that I'm surprised no one seems to have considered. It eliminates komi issues, basically. Here goes:
Two games are played, with colors switched in the second game. The scores - with no kom - from the two games for each player are added, the one with most points win. Note that, as with integer komi, draws are possible. As far as the draw issue, I've always thought that draws should be part of the game so that players of 'equal skill' - the only way to no this happens for sure is with programs, as many have said - draw, and so that there is no 'ultimate strategy' that always wins, only one that can always draw. However, it does require double the time. Fortunately, this doesn't matter much for tournaments, where several games are played. This method also tends to counteract, slightly, fluctuations in player's performance. Increasing the number of games and adding up scores to come up with a final outcome is one method to increase this 'fluctuation-damping', although it is inrelated to komi. Note that handicaps are still well defined: the player with handicap n starts with n stones of his color on the board, in both games, before black plays first. Any thoughts?
(First post! Yay!)
-aceofspades
Jared: One problem comes to mind immediately: about half of games end by resignation.
Mef: DGS Implements the "double game" feature which is basically the idea you describe, except I think that it just counts them as 2 separate games instead of 1 big game.
Another comment: I thought about this too. My proposition is that players play two games, and which one wins both of them, wins. Otherwise it could be considered as a draw. This kind of rating would be useful in ranking players, but may not result to pleasant gameplay. -- by meme on 051225.20
Jared, you're right in that resigning is a problem, but think about it like this: by resigning one game, you're really resigning both. So if you play the first game and lose by 20 points, you might still play the second game and try to catch up. If you lost by 60 points, you would probably resign. Of course, it's hard to decide whether to resign after the first game since you know nothing about the second game will go. One possible workaround, mainly suitable for tournaments or formal matches in general, is to play both games more or less simultaneously. That way you can combine your leads/deficits from both games to decide if you want to resign. Mef, I don't understand what you're saying - the whole point of what I'm suggesting is that the games are regarded as one big game. Maybe you mean that when the scores combine to give a win, that win is counted as 2 games won? Meme, the problem whith what you suggest is the same problem as with any 'segmentation' of games, like playing best-out-of-10 matches. Differences between players are amplified in best-of-10 matches, so a slightly better player will win much more often, and the match is only 'close' when both players are very closely matched. On the other hand, in the system you suggest, black's advantage would mean that in both games black would win unless white was significantly stronger, so the result of the 2-game-match would be 'pulled' toward 1-1, making it less useful. That's why a correct komi was needed - it makes the result of the game hingle only on the difference between player's skill, without weighting it. What I'm suggesting is similar, except that it is a way to overcome the problem of not knowing what the theoretically correct komi is. Thanks for the feedback! I'm going to try to put this mathematically. Let Bscore and Wscore be the scores in a perfectly played game. Then the definition of komi is komi=Bscore-Wscore. Th purpose of adding komi is that the scores are made equal so that any variations are due to player's mistakes/skill, and without requiring white to be especially stronger than black for the difference to show up, as happens when simply playing many seperate games with alternating colors. When, as I propose, two games are played with switched colors, each player's combined score is Bscore+Wscore under perfect play. Thus, the scores are equal, and variation is once again due to skill difference, and there is no 'pulling' of the results toward even. That is, the system I propose is entirely equivalent to komi but doesn't require one to estimate what the komi is. -aceofspades
I was just saying that the Dragon Go Server already has a "double game" feature where you play your opponent in 2 games simultaneously, one as black, one as white. However they don't add the results together, they just just treat them as 2 separate games. I also would like to echo Alex's comments about not truely eliminating the first move advantage, and instead are creating a go variant where margin of victory is included in your wins.
Alex: If the games aren't played simultaneously, you still haven't eliminated first move advantage. The player holding Black in the first game will still have an advantage (although smaller than with Black in a regular no-komi game), since he will win more than 50% of the time, and then know what his komi is when playing White in the second game - he will know how conservatively he can play and still win the match. The other player, however, has to play aggressively with White in the first game because he doesn't know how much use he'll get out of his first-move advantage in the second game.
Anyway, the biggest problem with this idea (if the purpose is to find a way to avoid using komi to play Go) is that you're no longer playing Go. One of the important aspects of Go is that a win is a win, regardless of the margin. Losing by 0.5, 80.5, resignation or time are all equivalent. Creating a variant in which your score is carried over between two or more games completely reverses the strategy. In Go, the player who is ahead will generally play more conservatively to avoid an upset, while the player who is behind will play aggressively to try to catch up, knowing that it makes no difference if he slips up and doubles his margin of loss. In your version, the player who is ahead (assuming he has a relatively thick position) will try to press that advantage to increase his lead and minimise the chance that the opponent will catch up in the second game, while a player with a small deficit will just try to minimise it, rather than trying to win, so as not to risk digging himself into too deep a hole to climb out of in the second game. Which is all fine if you're trying to create a new game, but not if you're trying to "fix" the komi issue in Go without changing the rest of the game.
Alex - I admit that what you say is true in terms of a probabilistic view of Go, but that's how humans play Go. In terms of Game theory, Go is still a perfectly sensible game if you add 1000 points to the other player's score, because you can still play to maximize your score. And the truth is that you wouldn't really care if 1000 points were added to your opponent's score if you knew that you had played the better moves. So in a sense, many games of go are really played only to maximize your score. As another example, what about a tesuji problem where weak group is under attack and the problem is to find the moves that save the most points? That's still Go.
-aceofspades
I've noticed that the higher a person gets in rank, the less this tends to be true.
Books such as Attack and Defense emphasize always knowing what the score is, because it can dramatically affect the way that one should play. For example, if I know I am behind (i.e., I counted) and don't think I can make up the difference in the endgame, I am more inclined to invade instead of reduce, or to go for a ko (or leave something as a ko) rather than settle it.
Because of this, it also isn't really a valid measure from the other side either. A player who starts attacking and playing aggressively--taking risks--in order to try and catch up will lose by quite a bit more than she might have otherwise. Further, what about the games that are won and lost by a liberty, rather than a few points here and there? Making the score count would change it into a very different game.
Finally, at a sufficiently high level (and pros still count as "humans") it isn't uncommon to "lock down a game." Look at how many games Lee Changho wins by 0.5--it is because, in Go--with the exception of some variants (gambling, team play where the scores are added and resigning is forbidden, etc), a win is a win is a win.
I know whenever I play I'm not playing to maximize my score. If I count and I'm ahead by 20, I will often give up 5-15 pts to simplfy the game and secure the win. Likewise, if I'm behind by 5 points I won't play solid to secure the 5 pt loss, I'll play moves that are more agressive and may lead to a 20 pt loss, but they at least give a chance for the win. Also tesuji problems are part of Tsumego (Restricted go) not Go. Tsumego is a go variant and it's also good proof that some go variants can be extremely helpful to playing regular Go.
PlatinumDragon: I have a similar idea to eliminating komi. In the first game, if A loses as black, then A lost the match. If A wins as black, then a second game plays with colors switch. In the second game, if A wins as white, then A wins the match. If A loses the match as white, then the game ends in a draw. At any point, a resign is a lost to the player that resign.
Round 1: A as black vs B as white.
If A wins, the proceed to round 2, else A lost the match.
Round 2: A as white vs B as black.
If A wins, then A wins the match, if A loses then the match is draw.
If a winner needs to be determined, then the sum score of the two games will determine the match.