Re: [computer-go] Anchor player
Le jeudi 17 janvier 2008, Don Dailey a écrit : Perfect! I will adjust the level so that it plays as strong as possible on CGOS without taking a risk of getting into time trouble on modest hardware. Then I can make Mogo the anchor player. Even if i love Mogo, and i am very impressed, i think it is a bad idea to use it as an anchor, as it is closed source. It can be used as a floating anchor (= a player always present, with no changes in settings), but i really think using GNU Go or one other open source program for the anchor is the best for the community. my 2 cents. Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor player
It's not real important for this test, but I think I will use gnugo-3.7.11 as the anchor and set it to 1800.0 ELO - which I think is fairly close to what it would do on CGOS. I will use level 10. - Don Alain Baeckeroot wrote: Le jeudi 17 janvier 2008, Don Dailey a écrit : Perfect! I will adjust the level so that it plays as strong as possible on CGOS without taking a risk of getting into time trouble on modest hardware. Then I can make Mogo the anchor player. Even if i love Mogo, and i am very impressed, i think it is a bad idea to use it as an anchor, as it is closed source. It can be used as a floating anchor (= a player always present, with no changes in settings), but i really think using GNU Go or one other open source program for the anchor is the best for the community. my 2 cents. Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
2006/12/26, Don Dailey [EMAIL PROTECTED]: There are many other ways to take advantage of your opponent in chess that I consider sound if applied in a very measured and careful way. None of them call for making truly unsound moves, especially when you consider that in a losing position, all moves are unsound in some sense.Now you are in a situation of risk management, you are looking for moves that give you the best chances of winning (a lost game) and usually, it requires a move that makes it the most difficult for your opponent. This is not quite the same as moves that make it easiest for you, which is what you look for in WON positions. There's one easy way I found to do this in Go. In handicap Go, if you're behind, set up a ko. :) Ko does complicate a game, and almost by definition, you will play it better and gain something. -- Seo Sanghyeon ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
On 12/27/06, Łukasz Lew [EMAIL PROTECTED] wrote: I agree :) What I wanted to ask is: Does there exists two final (no profitable move left) 9x9 board positions that their area score differ by one point ? Ah, sorry :) I believe there is, yes (that's what I was hinting at with almost). There must be a seki somewhere on the board though. -- nando ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
On 12/27/06, nando [EMAIL PROTECTED] wrote: On 12/27/06, Łukasz Lew [EMAIL PROTECTED] wrote: I agree :) What I wanted to ask is: Does there exists two final (no profitable move left) 9x9 board positions that their area score differ by one point ? Ah, sorry :) I believe there is, yes (that's what I was hinting at with almost). There must be a seki somewhere on the board though. Normal seki gives two neutral points, so it doesn't matter. Lukasz -- nando ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
On 12/27/06, Łukasz Lew [EMAIL PROTECTED] wrote: (...) Normal seki gives two neutral points, so it doesn't matter. I'm walking on increasingly thin ice (for me), but you're right, normal sekis shouldn't change things. Though, there are also beasts like this one: http://senseis.xmp.net/?path=LifeAndDeathpage=Seki#toc7 ... and I'm not sure how this would be counted under chinese rules. Anyway, I don't think this would cause any trouble to a 19x19 CGOS server. Or do you see a potential problem here ? -- nando ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
But not all of those are final (often dead stones remain on board). But one eye seki is an answer for me. Thanks, Lukasz On 12/27/06, Don Dailey [EMAIL PROTECTED] wrote: I extracted all the games from one month and uniq'd them, it looks like all results are possible: Forfeit B+ Forfeit W+ Illegal B+ Illegal W+ Resign B+ Resign W+ Time B+ Time W+ 0.5 B+ 0.5 W+ 1.5 B+ 1.5 W+ 2.5 B+ 2.5 W+ 3.5 B+ 3.5 W+ 4.5 B+ 4.5 W+ 5.5 B+ 5.5 W+ 6.5 B+ 6.5 W+ 7.5 B+ 7.5 W+ 8.5 B+ 8.5 W+ 9.5 B+ 9.5 W+ 10.5 B+ 10.5 W+ 11.5 B+ 11.5 W+ 12.5 B+ 12.5 W+ 13.5 B+ 13.5 W+ 14.5 B+ 14.5 W+ 15.5 B+ 15.5 W+ 16.5 B+ 16.5 W+ 17.5 B+ 17.5 W+ 18.5 B+ 18.5 W+ 19.5 B+ 19.5 W+ 20.5 B+ 20.5 W+ 21.5 B+ 21.5 W+ 22.5 B+ 22.5 W+ 23.5 B+ 23.5 W+ 24.5 B+ 24.5 W+ 25.5 B+ 25.5 W+ 26.5 B+ 26.5 W+ 27.5 B+ 27.5 W+ 28.5 B+ 28.5 W+ 29.5 B+ 29.5 W+ 30.5 B+ 30.5 W+ 31.5 B+ 31.5 W+ 32.5 B+ 32.5 W+ 33.5 B+ 33.5 W+ 34.5 B+ 34.5 W+ 35.5 B+ 35.5 W+ 36.5 B+ 36.5 W+ 37.5 B+ 37.5 W+ 38.5 B+ 38.5 W+ 39.5 B+ 39.5 W+ 40.5 B+ 40.5 W+ 41.5 B+ 41.5 W+ 42.5 B+ 42.5 W+ 43.5 B+ 44.5 B+ 44.5 W+ 45.5 B+ 45.5 W+ 46.5 B+ 46.5 W+ 47.5 B+ 47.5 W+ 48.5 B+ 48.5 W+ 49.5 B+ 49.5 W+ 50.5 B+ 50.5 W+ 51.5 B+ 51.5 W+ 52.5 B+ 52.5 W+ 53.5 B+ 53.5 W+ 54.5 B+ 54.5 W+ 55.5 B+ 55.5 W+ 56.5 W+ 57.5 B+ 57.5 W+ 58.5 B+ 58.5 W+ 59.5 B+ 59.5 W+ 60.5 B+ 60.5 W+ 61.5 B+ 61.5 W+ 62.5 B+ 62.5 W+ 63.5 B+ 63.5 W+ 64.5 B+ 64.5 W+ 65.5 B+ 66.5 B+ 66.5 W+ 67.5 B+ 68.5 B+ 68.5 W+ 69.5 B+ 69.5 W+ 70.5 B+ 70.5 W+ 71.5 W+ 72.5 W+ 73.5 B+ 74.5 W+ 75.5 W+ 76.5 W+ 77.5 W+ 78.5 W+ 79.5 W+ 80.5 W+ 81.5 W+ 82.5 W+ 83.5 W+ 84.5 W+ 85.5 W+ 88.5 W+ On Wed, 2006-12-27 at 14:56 +0100, Łukasz Lew wrote: On 12/27/06, nando [EMAIL PROTECTED] wrote: On 12/27/06, Łukasz Lew [EMAIL PROTECTED] wrote: I agree :) What I wanted to ask is: Does there exists two final (no profitable move left) 9x9 board positions that their area score differ by one point ? Ah, sorry :) I believe there is, yes (that's what I was hinting at with almost). There must be a seki somewhere on the board though. Normal seki gives two neutral points, so it doesn't matter. Lukasz -- nando ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Vlad Dumitrescu wrote: The best move may be a somewhat risky invasion - of course one has to assume the partner will not play perfectly, but everybody does that every time anyway, right? Otherwise nobody would have any hope to win and so nobody would play. I agree. That's easy for humans to understand. An unclear invasion is a blunder against a strong opponent but it is not against a weak one if you can trust your know-how to keep invading stones alive. But the point is how difficult it is for a computer to grasp subtlety. I think, but don't know, that MC will more naturally find the right measure of overplay than other approaches. It is a terribly bad idea to make a go program an adventurer. Therefore, prudent programs will always be underrated when they give handicap. They could achieve more if they underestimated their opponent assuming that handicap is given because they are stronger. Note that the best moment for overplay is the beginning, wait and see is not a good idea, the sharper the moyo is traced, the harder it is to invade. Later is too late. Jacques. PD Errata (in my previous post) 200 points = 1/(1+10^.50) = 0.3204 approx = 1/4 should be: 200 points = 1/(1+10^.50) = 0.2402 approx = 1/4 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player. ELO / handicap
Le lundi 25 décembre 2006 00:46, Don Dailey a écrit : On Sun, 2006-12-24 at 13:54 -0800, David Fotland wrote: There is no fixed relationship between ELO and handicap stones. Stronger players have less variation in their play, so a handicap stone is worth more ELO points for a stronger player than a weaker player. What you say is consistent with what I've heard from other sources. My understanding is that in ELO terms the ranks are compressed at the higher levels and spread out at lower levels. So there is less difference between 4 dan and 5 dan than 15 kyu and 16 kyu for instance. If I want to use ELO and also expect the handicaps to be fair, then I will need to account for this curve. Current KGS ranking seems very close to european ranking, so stats at http://gemma.ujf.cas.cz/~cieply/GO/statev.html can give usefull hint. GNU and other strong programs are in the range 10k-6k where the stats are rather regular, and rougly gives the follwowing winning percentage in even games (from more than 2 games) R + 1 R + 2 R + 3 R + 4 win% 44 403020 Equiv-ELO -43 -72 -149 -240 So a linear interpolation (even if it obviously not linear) gives approximately 50 ELO == 1 handi (for this range of strenght) On the web I see that some ELO based GO servers assume 100 ELO is 1 rank, and do exactly what I proposed, when they handicap they fold this into the ELO rating of the players for rating purposes. So taking 100 ELO for 1 k difference seems to be a good first guess, and gives slightly less handi than needed (this is good idea), and currently no one knows how bot ranking will look like ... Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player. High handi
Le lundi 25 décembre 2006 15:35, Jacques Basaldúa a écrit : I have seen (many times) GnuGo not being able to win a H7 game to an opponent more than 10 kyu weaker. That happens because it had to invade unclear positions. This is a feature of GNU Go :-) GNU Go has very small invasion capacity, and this is done on purpose, because it is too weak at making light plays, or move like escape-or-live. GNU is tuned for even games on 19X19, and so behaves far from optimal in high handicap games. Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
On 12/26/06, Don Dailey [EMAIL PROTECTED] wrote: Yes, the answer is that there is no gtp command available that defines whether handicap stones are also compensated or by how much. Just like there's no GTP command to define the ruleset. This compensation is 0 in japanese rules, N in chinese rules, N-1 in AGA rules, etc. So it seems more or less clearly defined by the ruleset used. CGOS 9x9 uses a modified Tromp-Taylor right ? What does this ruleset say about such compensation ? If it says nothing, let's just choose between N and N-1 (since a compensation is logical when area scoring is used), and add it to the set of input parameters that the operators have to pass off-line (out of the GTP stream). -- nando ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
On 12/26/06, Don Dailey [EMAIL PROTECTED] wrote: Yes, that's my plan.I'm going to use fixed handicap and 1 stone compensation per handicap stone. One question I have - is compensation normally given in the 1 stone case? I believe, no. Also, in the case of NO handicap, what komi is normally given in 19x19 Chinese? 6.5, 7.5 ??? It's 7.5 -- nando ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Hideki Kato wrote: In Nihon Kiin's ELO system(1), 1000 ELO is 1 rank, The Elo rating is based on two assumptions: a. The performance of each player in each game is a normally distributed random variable. b. All players performance have the same standard deviation. (This is controversial, and other rating systems modify this.) Anything else is arbitrary! Including how many Elo points produce a given probability. Elo *tradition* (from chess) has always used the convention: 100 points = 1/(1+10^.25) = 0.3599 approx = 1/3 200 points = 1/(1+10^.50) = 0.3204 approx = 1/4 1000 Elo points give a probability of 1/317 It is obvious that Nihon Kiin's use a different scale (may be 1000 Nihon Kiin's = 100 traditional). On the handicap subject: I am very happy to have a 19x19 server either with or without handicap, so I welcome it as it is. Nevertheless, I have certain experience (not with MC) of computer go with handicap and I can tell: Waiting for the opponent to blunder is only a good strategy if the handicap is lower than it should. E.g. 7 kyu difference Handi 3. If the handicap approaches its real value, that does not work. I have seen (many times) GnuGo not being able to win a H7 game to an opponent more than 10 kyu weaker. That happens because it had to invade unclear positions. The more the invasion is postponed, the worse. The weaker player simply does defensive uninteresting play and so does the stronger player (with better yose, but that's not enough). If I (manually) use two or three turns just to invade, GnuGo tries to save the invading stones and that's more than enough to win the game. As I said before, its a different game and the more accurate you determine the handicap, the worse. If at all, handicap should always be underestimated by a factor of 1/2. Handicap is used to make the game interesting enough to white (but usually white still wins) to honor a lower player with a learning game. I hope weaker bots will learn at lot from the games played against the stronger. ;-) Jacques. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
On 12/25/06, Jacques Basaldúa [EMAIL PROTECTED] wrote: Hideki Kato wrote: Nevertheless, I have certain experience (not with MC) of computer go with handicap and I can tell: Waiting for the opponent to blunder is only a good strategy if the handicap is lower than it should. E.g. 7 kyu difference Handi 3. If the handicap approaches its real value, that does not work. I have seen (many times) GnuGo not being able to win a H7 game to an opponent more than 10 kyu weaker. That happens because it had to invade unclear positions. The more the invasion is postponed, the worse. The weaker player simply does defensive uninteresting play and so does the stronger player (with better yose, but that's not enough). If I (manually) use two or three turns just to invade, GnuGo tries to save the invading stones and that's more than enough to win the game. Hi, This depends on your definitions. If the position is unclear and GnuGo doesn't invade, then I'd say *that* is a blunder (especially when being behind). The idea is to play the best move available, and let the weaker player make suboptimal ones -- in the end, if the handicap is correct, the net result should be zero and the result would be the same as in an even game with an equal partner. The best move may be a somewhat risky invasion - of course one has to assume the partner will not play perfectly, but everybody does that every time anyway, right? Otherwise nobody would have any hope to win and so nobody would play ;-) best regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Hi Hideki, I think what I will do is use ELO and a simple formula for determining handicap. The formula will impose a slight curve on the value of a handicap stone, it will slightly increase with each ELO point. In other words a stronger player will benefit more from having an extra stone and the handicap will be chosen appropriately. Of course the formula will be an assumption. I will either build something in to the server to make gentle modifications over time, or I will manually adjust the parameters from time to time based on the data collected from the server. It will be easy to tell if the handicaps are too aggressive or too conservative after a lot of data is collected. The initial formula will assume the stronger players on 19x19 CGOS need 1 handicap stone to overcome 100 ELO points. This seems to be fairly standard in servers and I think it's probably a good starting point.Since computer programs do not represent very strong players at 19x19 I don't think there is a great deal of rank compression at these levels, so I can imagine that this will be a reasonable starting point for the stronger players. Of course 100 ELO per stone will probably not work well with the really weak players (and may even be wrong for the strong computer players) so one way or another we will have to converge on a formula that tries to be as fair as possible. There seems to be a large range of computer playing skill on CGOS, from zero ELO to 2200 without any large gaps.This may look different at 19x19, we shall see. I think our formula will require 2 constants that can be adjusted to control the shape of the curve. I will come up with something but I will be happy to take suggestions too. Like everything else that has to do with ratings, rankings, handicaps, etc this is all an estimate and will never be 100% perfect. But I think we can make the attempt to fit the data according to the results to be as fair as possible. - Don On Mon, 2006-12-25 at 14:35 +, Jacques Basaldúa wrote: Hideki Kato wrote: In Nihon Kiin's ELO system(1), 1000 ELO is 1 rank, The Elo rating is based on two assumptions: a. The performance of each player in each game is a normally distributed random variable. b. All players performance have the same standard deviation. (This is controversial, and other rating systems modify this.) Anything else is arbitrary! Including how many Elo points produce a given probability. Elo *tradition* (from chess) has always used the convention: 100 points = 1/(1+10^.25) = 0.3599 approx = 1/3 200 points = 1/(1+10^.50) = 0.3204 approx = 1/4 1000 Elo points give a probability of 1/317 It is obvious that Nihon Kiin's use a different scale (may be 1000 Nihon Kiin's = 100 traditional). On the handicap subject: I am very happy to have a 19x19 server either with or without handicap, so I welcome it as it is. Nevertheless, I have certain experience (not with MC) of computer go with handicap and I can tell: Waiting for the opponent to blunder is only a good strategy if the handicap is lower than it should. E.g. 7 kyu difference Handi 3. If the handicap approaches its real value, that does not work. I have seen (many times) GnuGo not being able to win a H7 game to an opponent more than 10 kyu weaker. That happens because it had to invade unclear positions. The more the invasion is postponed, the worse. The weaker player simply does defensive uninteresting play and so does the stronger player (with better yose, but that's not enough). If I (manually) use two or three turns just to invade, GnuGo tries to save the invading stones and that's more than enough to win the game. As I said before, its a different game and the more accurate you determine the handicap, the worse. If at all, handicap should always be underestimated by a factor of 1/2. Handicap is used to make the game interesting enough to white (but usually white still wins) to honor a lower player with a learning game. I hope weaker bots will learn at lot from the games played against the stronger. ;-) Jacques. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
I was always taught in Chess to play the board, not the player. But in principle this is wrong if your goal is to increase your chances of winning the game. The problem with playing your opponent is that if you don't know the proper technique for doing this, it will distract you from the real game.For weaker players, it's enough to just play the board, anything beyond this will hurt their play. What most people don't understand is that playing your opponent doesn't mean that you suddenly start playing risky unsound moves. Playing your opponent is a very careful and controlled process that should not involve a radical change of playing style. It involves waiting for mistakes, not making them. You will try to encourage these mistakes, but not at great risk to yourself. Also, you must never underestimate your weaker opponent. I mention this because there have been a few posts that imply that you should play normally, change nothing, and just wait for the mistake. In fact, this is almost correct, most of the moves should be like this.If you don't know what you are doing ALL the moves should be like this. But if you want to actually maximize your winning chances, you need to be more sophisticated than this. At least this applies in Chess. I'm sure this must apply to GO too. When books and experts say don't play the opponent they are giving beginner advice. Most beginners can't handle this. It should be done in very carefully measured ways.It's obviously counter-productive to start playing high risk moves and throw soundness out the window.But that doesn't mean there is nothing you can do to take advantage of a weaker opponent in a losing position. For most people, the advice to just play the board is going to protect them. Playing the opponent is a skill and weaker (chess players) screw up big time when attempting to do this. I'll give one example from chess: What do you do when your opponent is in time-trouble? How do you play the opponent and capitalize on this? The knee-jerk reaction is to play extra quickly, to deprive him of thinking on your time. This is foolish. Your opponent will be at his best with the extra adrenaline kick. If you play fast you will be at your worst. The adrenaline rush will wear out your opponent, so in such a situation it is better to play normally or even EXTRA SLOW. In fact, if I see my opponent is excited, I take my sweet time, forcing him to stay at attention a very long time. If I have a choice between complicated and simple positions, I take the complicated position IF and ONLY IF I am totally comfortable with it. After all I have more time to figure it out. I won't play unsoundly just to get a complicated position though. The better players are not really that handicapped when under time pressure - they remain highly focused and do not let themselves get too excited - no point in helping them along with rash moves.YOU are the one that must remain calm and stay 100 percent focused. I did not have to lose very many games that way to figure this out. There are many other ways to take advantage of your opponent in chess that I consider sound if applied in a very measured and careful way. None of them call for making truly unsound moves, especially when you consider that in a losing position, all moves are unsound in some sense.Now you are in a situation of risk management, you are looking for moves that give you the best chances of winning (a lost game) and usually, it requires a move that makes it the most difficult for your opponent. This is not quite the same as moves that make it easiest for you, which is what you look for in WON positions. Weaker players cannot do this. They are best just sticking with the style that is most comfortable for them. - Don On Mon, 2006-12-25 at 15:23 +, Vlad Dumitrescu wrote: On 12/25/06, Jacques Basaldúa [EMAIL PROTECTED] wrote: Hideki Kato wrote: Nevertheless, I have certain experience (not with MC) of computer go with handicap and I can tell: Waiting for the opponent to blunder is only a good strategy if the handicap is lower than it should. E.g. 7 kyu difference Handi 3. If the handicap approaches its real value, that does not work. I have seen (many times) GnuGo not being able to win a H7 game to an opponent more than 10 kyu weaker. That happens because it had to invade unclear positions. The more the invasion is postponed, the worse. The weaker player simply does defensive uninteresting play and so does the stronger player (with better yose, but that's not enough). If I (manually) use two or three turns just to invade, GnuGo tries to save the invading stones and that's more than enough to win the game. Hi, This depends on your definitions. If the position is unclear and GnuGo doesn't invade, then I'd say *that* is a blunder (especially when being
Re: [computer-go] Anchor Player
2006/12/25, Don Dailey [EMAIL PROTECTED]: On Sun, 2006-12-24 at 13:54 -0800, David Fotland wrote: There is no fixed relationship between ELO and handicap stones. Stronger players have less variation in their play, so a handicap stone is worth more ELO points for a stronger player than a weaker player. What you say is consistent with what I've heard from other sources. My understanding is that in ELO terms the ranks are compressed at the higher levels and spread out at lower levels. So there is less difference between 4 dan and 5 dan than 15 kyu and 16 kyu for instance. I think it's exactly the opposite. The difference between 4 dan and 5 dan is one stone, but more ELO than between 15 and 16k (also one stone). Andrés ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Are you sure about this? Here is what I've seen on Wikipedia but I've also seen this before from other sources: Another departure from tradition is that ELO ratings are calibrated by winning percentage, not by stone handicaps. An extra handicap stone has much less influence on winning percentage at a low level of play than at a high level of play. Therefore, from the perspective of ELO ratings, traditional ranks are too spread out at the low level and too compressed at a high level. To put it another way, a 6-dan player has a much better chance of beating a 5-dan player than a 15-kyu player has of beating a 16-kyu player, so the ELO system must conclude either that the top players need to be further apart in rating than 100 points, or the bottom players need to be closer in rating than 100 points. - Don On Mon, 2006-12-25 at 20:23 +0100, Andrés Domínguez wrote: 2006/12/25, Don Dailey [EMAIL PROTECTED]: On Sun, 2006-12-24 at 13:54 -0800, David Fotland wrote: There is no fixed relationship between ELO and handicap stones. Stronger players have less variation in their play, so a handicap stone is worth more ELO points for a stronger player than a weaker player. What you say is consistent with what I've heard from other sources. My understanding is that in ELO terms the ranks are compressed at the higher levels and spread out at lower levels. So there is less difference between 4 dan and 5 dan than 15 kyu and 16 kyu for instance. I think it's exactly the opposite. The difference between 4 dan and 5 dan is one stone, but more ELO than between 15 and 16k (also one stone). Andrés ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Anchor Player
There is no fixed relationship between ELO and handicap stones. Stronger players have less variation in their play, so a handicap stone is worth more ELO points for a stronger player than a weaker player. David -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Don Dailey Sent: Friday, December 22, 2006 2:04 PM To: computer-go Subject: Re: [computer-go] Anchor Player So really, what I want to be able to do is: 1. Use the ELO rating system. 2. Determine how many ELO points 1 stone handicap is worth. 3. 2 stones are worth 4. 3 stones are worth, ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Quoting Christian Nilsson [EMAIL PROTECTED]: Yes, in chinese rules you need to compensate white for the extra area you gain from the actual stones. The handicap is only meant to be the extra strength/stability. One can of course ignore this for the server. I just wanted to make sure all programs use the same rules. I don't know what the tromp-taylor rules says about it. Please, do use this compensation because KGS use it (correct me if I am wrong) and it would be very silly to make all programs buggy on at least on server. The compensation is there to make japanese and chinese give the same result in handicap games, which is very important to make the game consistent for all players no matter what background they have. -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Based on some research I've done, it does make some sense to give some compensation for handicap stones, because it makes it match Japanese and without it, the kyu system is not balanced. I have doubts that it's perfectly balanced anyway, but that's a different subject. So I think we will include this in our handicap server. I want to make it as unambiguous as possible. There are some ways to handle this: 1. The program must know how to make the compensation. OR 2. The server adds the stones to komi and the program pretends there is no such thing as compensation - it's just built in to komi. OR 3. We have a GTP command to handle it. (presumably there is no specific gtp command to deal with it. OR any suggestions? The basic idea (which matches KGS) is that if you give a 4 stone handicap, white gets 4 stones of compensation, as if komi has increased by 4. - Don On Sat, 2006-12-23 at 09:28 +0100, Magnus Persson wrote: Quoting Christian Nilsson [EMAIL PROTECTED]: Yes, in chinese rules you need to compensate white for the extra area you gain from the actual stones. The handicap is only meant to be the extra strength/stability. One can of course ignore this for the server. I just wanted to make sure all programs use the same rules. I don't know what the tromp-taylor rules says about it. Please, do use this compensation because KGS use it (correct me if I am wrong) and it would be very silly to make all programs buggy on at least on server. The compensation is there to make japanese and chinese give the same result in handicap games, which is very important to make the game consistent for all players no matter what background they have. -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
On Sat, 2006-12-23 at 20:20 +0100, Łukasz Lew wrote: On 12/23/06, Don Dailey [EMAIL PROTECTED] wrote: Based on some research I've done, it does make some sense to give some compensation for handicap stones, because it makes it match Japanese and without it, the kyu system is not balanced. I have doubts that it's perfectly balanced anyway, but that's a different subject. So I think we will include this in our handicap server. I want to make it as unambiguous as possible. There are some ways to handle this: 1. The program must know how to make the compensation. OR 2. The server adds the stones to komi and the program pretends there is no such thing as compensation - it's just built in to komi. OR 3. We have a GTP command to handle it. (presumably there is no specific gtp command to deal with it. OR any suggestions? AGA rules solve the problem gracefully. http://www.cs.cmu.edu/~wjh/go/rules/AGA.concise.html I suggest to use them. Łukasz, There is nothing on that site that addresses this issue, in fact the site seems geared towards human play and not computer play. Not only that, but the rules are different than CGOS uses. In fact the rules don't even specify a standard, lot's of options like which kind of scoring system to use, etc. There is a paragraph on scoring disputes and how to handle them, but nothing that I would consider graceful and elegant. The problem I'm trying to address is how to communicate, or not communicate to the programs what handicap system is in place, and whether handicap stones are compensated.I'm not trying to figure out which handicap system to use, I've already decided to go with the KGS system that Magnus is advocating. The simplest thing is to just explain it on a web page, but there is no explicit way to tell the programs that white is being compensated (or not) for the handicap stones and that bothers me. - Don Łukasz Lew The basic idea (which matches KGS) is that if you give a 4 stone handicap, white gets 4 stones of compensation, as if komi has increased by 4. - Don On Sat, 2006-12-23 at 09:28 +0100, Magnus Persson wrote: Quoting Christian Nilsson [EMAIL PROTECTED]: Yes, in chinese rules you need to compensate white for the extra area you gain from the actual stones. The handicap is only meant to be the extra strength/stability. One can of course ignore this for the server. I just wanted to make sure all programs use the same rules. I don't know what the tromp-taylor rules says about it. Please, do use this compensation because KGS use it (correct me if I am wrong) and it would be very silly to make all programs buggy on at least on server. The compensation is there to make japanese and chinese give the same result in handicap games, which is very important to make the game consistent for all players no matter what background they have. -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Quoting Don Dailey [EMAIL PROTECTED]: The simplest thing is to just explain it on a web page, but there is no explicit way to tell the programs that white is being compensated (or not) for the handicap stones and that bothers me. The first step is to inform future programmers of the compensation. KGS uses a special command for playing and placing free handicap. Right now I do not remember if this is part of the GTP protocol or if it is an extension of KGS. The only hard problem I can see would be if the server would fake handicap by playing black moves with white passes. Then it would be impossible to know for sure if it is a handicap game or not. But as long as there are explicit commands for the handicap given it is not hard to implement and it would also work later because as far as I know all important servers and programs uses the compensation. -Merry christmas! Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
I think what I will do is see if there is an existing gtp command, if not I will see if there is a kgs extension for it - if there is I will imitate it with a cgos extension. If a program doesn't honor the extension I'll just document how it works and what to expect. I'm not going to fake handicap - there is already GTP command in place for that. Will have to slightly extend the CGOS client to handle it though. Will make a web page that clearly explains how everything works. - Don On Sat, 2006-12-23 at 22:37 +0100, Magnus Persson wrote: Quoting Don Dailey [EMAIL PROTECTED]: The simplest thing is to just explain it on a web page, but there is no explicit way to tell the programs that white is being compensated (or not) for the handicap stones and that bothers me. The first step is to inform future programmers of the compensation. KGS uses a special command for playing and placing free handicap. Right now I do not remember if this is part of the GTP protocol or if it is an extension of KGS. The only hard problem I can see would be if the server would fake handicap by playing black moves with white passes. Then it would be impossible to know for sure if it is a handicap game or not. But as long as there are explicit commands for the handicap given it is not hard to implement and it would also work later because as far as I know all important servers and programs uses the compensation. -Merry christmas! Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
On 12/23/06, Don Dailey [EMAIL PROTECTED] wrote: On Sat, 2006-12-23 at 20:20 +0100, Łukasz Lew wrote: On 12/23/06, Don Dailey [EMAIL PROTECTED] wrote: Based on some research I've done, it does make some sense to give some compensation for handicap stones, because it makes it match Japanese and without it, the kyu system is not balanced. I have doubts that it's perfectly balanced anyway, but that's a different subject. So I think we will include this in our handicap server. I want to make it as unambiguous as possible. There are some ways to handle this: 1. The program must know how to make the compensation. OR 2. The server adds the stones to komi and the program pretends there is no such thing as compensation - it's just built in to komi. OR 3. We have a GTP command to handle it. (presumably there is no specific gtp command to deal with it. OR any suggestions? AGA rules solve the problem gracefully. http://www.cs.cmu.edu/~wjh/go/rules/AGA.concise.html I suggest to use them. Łukasz, There is nothing on that site that addresses this issue, in fact the site seems geared towards human play and not computer play. Not only that, but the rules are different than CGOS uses. In fact the rules don't even specify a standard, lot's of options like which kind of scoring system to use, etc. There is a paragraph on scoring disputes and how to handle them, but nothing that I would consider graceful and elegant. The problem I'm trying to address is how to communicate, or not communicate to the programs what handicap system is in place, and whether handicap stones are compensated.I'm not trying to figure out which handicap system to use, I've already decided to go with the KGS system that Magnus is advocating. The simplest thing is to just explain it on a web page, but there is no explicit way to tell the programs that white is being compensated (or not) for the handicap stones and that bothers me. - Don Don, I will cite it here: If the players have agreed to use area counting to score the game (Rule 12), White receives an additional point of compensation for each Black handicap stone after the first. So AGA rules just do compensation. What I consider graceful is that AGA rules support both area and territory scoring, giving the same result. (last paragraph from the web page) Best Regards, Łukasz Łukasz Lew The basic idea (which matches KGS) is that if you give a 4 stone handicap, white gets 4 stones of compensation, as if komi has increased by 4. - Don On Sat, 2006-12-23 at 09:28 +0100, Magnus Persson wrote: Quoting Christian Nilsson [EMAIL PROTECTED]: Yes, in chinese rules you need to compensate white for the extra area you gain from the actual stones. The handicap is only meant to be the extra strength/stability. One can of course ignore this for the server. I just wanted to make sure all programs use the same rules. I don't know what the tromp-taylor rules says about it. Please, do use this compensation because KGS use it (correct me if I am wrong) and it would be very silly to make all programs buggy on at least on server. The compensation is there to make japanese and chinese give the same result in handicap games, which is very important to make the game consistent for all players no matter what background they have. -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Hi, On 12/22/06, Stuart A. Yeates [EMAIL PROTECTED] wrote: On 12/21/06, Jacques Basaldúa [EMAIL PROTECTED] wrote: Handicap play is a *different* problem. The rules of go include rules for handicapping. It seems to me that this implies that a complete solution for the game of go must include the ability to play such games. Yes, of course. But is it that difficult? The goal would 'just' have to change from winning to getting the best possible result. Now if one has already solved the game for the former goal, it should be trivial to adapt it for the latter, right? As a matter of fact, after solving the game for any goal, almost any computer science related matter would become rather trivial, I think :-) I.e. if the NP complete problems are solved, only easy ones remain! best regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Hi Don, On 12/22/06, Don Dailey [EMAIL PROTECTED] wrote: It's easy to adapt monte carlo programs to have the goal of trying to win as much space or territory as possible but many of us have studied this as see that it seriously weakens monte carlo programs. My (jokingly serious) point was that if you succeed solving the normal game of Go, fixing it for this additional constraing should be trivial (i.e. possibly only some 6 to 8 orders of magnitude simpler) But this is not the real problem. It seems that the handicap system is not reasonable in general for computers. [...] It seems that playing the best move possible (best in the sense of maximizing your territory gain) is not the best strategy when playing a handicap game. You literally have to play foolishly in order to dupe your opponent into losing. I would beg to partially disagree. The above is true if giving handicap to a player of equal strength, or at least stronger than the handicap would be fair for. IMHO if I give handicap it is because the other player is weaker, so I don't *have* to play foolishly - he will make mistakes that I can see and exploit. If I still can't win, it means the handicap should be lowered... This is the strategy that one uses even in even games, right? One plays what one thinks is best given the position, and if the opponent's reply is less than optimal one tries to punish it (with various degrees of success, but that's another issue :-)) Best regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
This is the strategy that one uses even in even games, right? One plays what one thinks is best given the position, and if the opponent's reply is less than optimal one tries to punish it (with various degrees of success, but that's another issue :-)) It's the strategy in even games, but not in handicap games. Let me put it like this. For simplification, in an even position we assume that if you play the objectively best move you will win the game (not entirely true but for all practical purposes we can assume this is true.)In other words, the quality of your opponents moves do not need to concern you. You have a simple strategy, just play the best move irregardless of what your opponent does. As the handicap grows, you become far more dependent on how your opponent plays. In fact, playing the very best move is a losing proposition from a game theoretic point of view, because there is no best move - they all lose! In fact, this is the crux of the matter. You say play the best move but in a losing position that is meaningless.This isn't just theoretical, it's practical, ALL moves are losing moves. So it becomes far more important to play the opponent, not the board. All your hopes and dreams depend on your opponent, not the brilliancy of your moves (all of which lose.) So it makes a great deal of sense to understand your opponent and to play in such a way that your opponent is more likely to go wrong. I'm not aware of any computers that think in these terms. However, humans do! I remember seeing a game annotated where a good player beat a program with some huge number of handicap stones. The annotations made it very clear that the human player was far more concerned with his opponent than the board. I'm fairly confident that in low handicap games where there is not a great deal of strength difference between players, this can be ignored without too many side effects. The same issues I describe exist, but we may be able to safely ignore them. I can't say that for sure since I am not a strong player. - Don On Fri, 2006-12-22 at 15:33 +0100, Vlad Dumitrescu wrote: Hi Don, On 12/22/06, Don Dailey [EMAIL PROTECTED] wrote: It's easy to adapt monte carlo programs to have the goal of trying to win as much space or territory as possible but many of us have studied this as see that it seriously weakens monte carlo programs. My (jokingly serious) point was that if you succeed solving the normal game of Go, fixing it for this additional constraing should be trivial (i.e. possibly only some 6 to 8 orders of magnitude simpler) But this is not the real problem. It seems that the handicap system is not reasonable in general for computers. [...] It seems that playing the best move possible (best in the sense of maximizing your territory gain) is not the best strategy when playing a handicap game. You literally have to play foolishly in order to dupe your opponent into losing. I would beg to partially disagree. The above is true if giving handicap to a player of equal strength, or at least stronger than the handicap would be fair for. IMHO if I give handicap it is because the other player is weaker, so I don't *have* to play foolishly - he will make mistakes that I can see and exploit. If I still can't win, it means the handicap should be lowered... This is the strategy that one uses even in even games, right? One plays what one thinks is best given the position, and if the opponent's reply is less than optimal one tries to punish it (with various degrees of success, but that's another issue :-)) Best regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Hi Steve, What you fail to take into considerations is that a monte/carlo player may ruin it's chances before the weaker player has a chance to play a bad move. The monte carlo player sees all moves as losing and will play almost randomly. In botnoids game against mogo, once mogo achieved a hopelessly won position, it conceded a lot of territory. Mogo only won by 1 or 2 stones because winning big either conflicted with the goal of winning, or at best didn't matter. But the converse also applies. Had Mogo been in a hopelessly LOST position, it would not have cared about saving face and a random move again would have sufficed (in a lost position only losing moves exist, so one is as good as another.) In a high handicap game, a monte carlo program is likely to play the first few move randomly. Statistically they won't be able to see how C3 is any better than A19 and so they will inadvertently give the weaker opponent the win. Presumably the handicap gives both player equal chances, so the stronger player, despite his superiority, cannot be doing this. - Don On Fri, 2006-12-22 at 06:46 -0800, steve uurtamo wrote: IMHO if I give handicap it is because the other player is weaker, so I don't *have* to play foolishly - he will make mistakes that I can see and exploit. If I still can't win, it means the handicap should be lowered... and any go program would operate the same way. it would look hopeless at the first move (as it does to any white player who is giving a handicap), but as soon as a mistake was made, white would exploit it and his probability to win (or whatever measure he's using) would increase. yes, it would look entirely desperate until those inefficient moves were played by his opponent, but it always does to humans as well. in a 9-stone game, you can expect a move *that you, as white, can see is inefficient* in, say, the first 5 moves. in a 2-stone game, perhaps the first 50. to a probability of win program, this would just look like a massive jump in white's probability to win. which is good news for white. depending upon when you see them and how bad these moves are, the handicap should be effectively negated before or just after the start of yose. MC (for instance) shouldn't have any trouble with this. s. __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Don Dailey wrote: Hi Steve, What you fail to take into considerations is that a monte/carlo player may ruin it's chances before the weaker player has a chance to play a bad move. The monte carlo player sees all moves as losing and will play almost randomly. I don't agree. Here is the winning percentage I get with Crazy Stone at various handicaps, with a komi of 0.5, over 1 random simulations: 9 Stones: 0.74 8 Stones: 0.73 7 Stones: 0.69 6 Stones: 0.67 5 Stones: 0.63 4 Stones: 0.61 3 Stones: 0.57 2 Stones: 0.54 My program still plays reasonable moves at these winning rates. I tend to believe MC programs would handle handicap better than pure territory-based programs, because they know how to play safe when they are ahead, and risky when behind. If my program is much stronger than its opponent, then it will not play blunders that the opponent can easily take advantage of, whatever the handicap. This being said, I don't believe my program can give handicap to any other on 19x19 ;-) Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Thanks for sending the statistics. I'll try them out later on my programs too. There is only 1 way to resolve this - maybe we should test it out on a 19x19 handicap server. We can play a few weeks and then take a look at the statistics later. I predict that gnugo will perform better on handicap games relative to non-handicap games and monte carlo players. I agree that monte carlo program will do very well when they are given the advantage. My fear is that the weak monte carlo program will do better than they should against the stronger ones when give the advantage of handicap games. I will also predict that your 9 stone 0.74 score will very quickly degrade to close to 1.0 after a few moves.9 stones is a dead won game and the score is far from accurate at 0.74. Can you try playing a couple of reasonable moves against crazy stone at this handicap and see what happens to the score? - Don On Fri, 2006-12-22 at 17:25 +0100, Rémi Coulom wrote: Don Dailey wrote: Hi Steve, What you fail to take into considerations is that a monte/carlo player may ruin it's chances before the weaker player has a chance to play a bad move. The monte carlo player sees all moves as losing and will play almost randomly. I don't agree. Here is the winning percentage I get with Crazy Stone at various handicaps, with a komi of 0.5, over 1 random simulations: 9 Stones: 0.74 8 Stones: 0.73 7 Stones: 0.69 6 Stones: 0.67 5 Stones: 0.63 4 Stones: 0.61 3 Stones: 0.57 2 Stones: 0.54 My program still plays reasonable moves at these winning rates. I tend to believe MC programs would handle handicap better than pure territory-based programs, because they know how to play safe when they are ahead, and risky when behind. If my program is much stronger than its opponent, then it will not play blunders that the opponent can easily take advantage of, whatever the handicap. This being said, I don't believe my program can give handicap to any other on 19x19 ;-) Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Quoting [EMAIL PROTECTED]: on 9x9 boards. To make a long story short, I didn't see any evidence that this algorithm is fundamentally disadvantaged in handicap games. In fact, I agree with Remi's view that it is particularly *well* suited to handicap games compared to territory based algorithms. When it finds itself behind, it goes for the swindle. Against an equal opponent, that's obnoxious and futile. Against a weaker foe, it's wise. I have to add to this. My opinion about Valkyrias play as white with handicap on 9x9 is that it plays excellent handicap go (given its strength at even). In my view I do not have to change anything. But this is of course just my impression. It is true that MC-programs plays randomly in the end of the game but in the opening the handicap stones are just a burden and does not really make it play random. -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
There's also the small issue of the compensation given to white because of the extra black stones on the board. Setting a modified komi would break (MC-)programs with an internal rule for it. Not setting it would break those who does not use that rule. How is this compensation handled by the various programs on cgos, if at all? Check http://www.britgo.org/rules/compare.html#comp if you don't know what I'm talking about.. //Christian On 12/22/06, Don Dailey [EMAIL PROTECTED] wrote: Ok, Well I'm inclined to go with the majority which seems to have turned around from the last time I polled. Now the question: How to set it up? Here are the options: 1. Use GTP handicap commands to set up game. 2. Send the appropriate pass commands to get the initial setup to accommodate programs that have not implemented handicap. 3. Do both - send handicap to programs that can handle it, otherwise guide them through it by sending play commands with passes. In any case, I would make the game records (SGF) look correct, doing whatever that takes. - Don On Fri, 2006-12-22 at 20:48 +0100, Magnus Persson wrote: Quoting [EMAIL PROTECTED]: on 9x9 boards. To make a long story short, I didn't see any evidence that this algorithm is fundamentally disadvantaged in handicap games. In fact, I agree with Remi's view that it is particularly *well* suited to handicap games compared to territory based algorithms. When it finds itself behind, it goes for the swindle. Against an equal opponent, that's obnoxious and futile. Against a weaker foe, it's wise. I have to add to this. My opinion about Valkyrias play as white with handicap on 9x9 is that it plays excellent handicap go (given its strength at even). In my view I do not have to change anything. But this is of course just my impression. It is true that MC-programs plays randomly in the end of the game but in the opening the handicap stones are just a burden and does not really make it play random. -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
I'm trying to figure this out. If you get a 9 stone handicap, you have to give back those 9 stones? So a 9 stone handicap is not quite as much as it seems although it's still pretty good. - Don On Fri, 2006-12-22 at 21:24 +0100, Christian Nilsson wrote: There's also the small issue of the compensation given to white because of the extra black stones on the board. Setting a modified komi would break (MC-)programs with an internal rule for it. Not setting it would break those who does not use that rule. How is this compensation handled by the various programs on cgos, if at all? Check http://www.britgo.org/rules/compare.html#comp if you don't know what I'm talking about.. //Christian On 12/22/06, Don Dailey [EMAIL PROTECTED] wrote: Ok, Well I'm inclined to go with the majority which seems to have turned around from the last time I polled. Now the question: How to set it up? Here are the options: 1. Use GTP handicap commands to set up game. 2. Send the appropriate pass commands to get the initial setup to accommodate programs that have not implemented handicap. 3. Do both - send handicap to programs that can handle it, otherwise guide them through it by sending play commands with passes. In any case, I would make the game records (SGF) look correct, doing whatever that takes. - Don On Fri, 2006-12-22 at 20:48 +0100, Magnus Persson wrote: Quoting [EMAIL PROTECTED]: on 9x9 boards. To make a long story short, I didn't see any evidence that this algorithm is fundamentally disadvantaged in handicap games. In fact, I agree with Remi's view that it is particularly *well* suited to handicap games compared to territory based algorithms. When it finds itself behind, it goes for the swindle. Against an equal opponent, that's obnoxious and futile. Against a weaker foe, it's wise. I have to add to this. My opinion about Valkyrias play as white with handicap on 9x9 is that it plays excellent handicap go (given its strength at even). In my view I do not have to change anything. But this is of course just my impression. It is true that MC-programs plays randomly in the end of the game but in the opening the handicap stones are just a burden and does not really make it play random. -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Yes, in chinese rules you need to compensate white for the extra area you gain from the actual stones. The handicap is only meant to be the extra strength/stability. One can of course ignore this for the server. I just wanted to make sure all programs use the same rules. I don't know what the tromp-taylor rules says about it. On 12/22/06, Don Dailey [EMAIL PROTECTED] wrote: I'm trying to figure this out. If you get a 9 stone handicap, you have to give back those 9 stones? So a 9 stone handicap is not quite as much as it seems although it's still pretty good. - Don On Fri, 2006-12-22 at 21:24 +0100, Christian Nilsson wrote: There's also the small issue of the compensation given to white because of the extra black stones on the board. Setting a modified komi would break (MC-)programs with an internal rule for it. Not setting it would break those who does not use that rule. How is this compensation handled by the various programs on cgos, if at all? Check http://www.britgo.org/rules/compare.html#comp if you don't know what I'm talking about.. //Christian On 12/22/06, Don Dailey [EMAIL PROTECTED] wrote: Ok, Well I'm inclined to go with the majority which seems to have turned around from the last time I polled. Now the question: How to set it up? Here are the options: 1. Use GTP handicap commands to set up game. 2. Send the appropriate pass commands to get the initial setup to accommodate programs that have not implemented handicap. 3. Do both - send handicap to programs that can handle it, otherwise guide them through it by sending play commands with passes. In any case, I would make the game records (SGF) look correct, doing whatever that takes. - Don On Fri, 2006-12-22 at 20:48 +0100, Magnus Persson wrote: Quoting [EMAIL PROTECTED]: on 9x9 boards. To make a long story short, I didn't see any evidence that this algorithm is fundamentally disadvantaged in handicap games. In fact, I agree with Remi's view that it is particularly *well* suited to handicap games compared to territory based algorithms. When it finds itself behind, it goes for the swindle. Against an equal opponent, that's obnoxious and futile. Against a weaker foe, it's wise. I have to add to this. My opinion about Valkyrias play as white with handicap on 9x9 is that it plays excellent handicap go (given its strength at even). In my view I do not have to change anything. But this is of course just my impression. It is true that MC-programs plays randomly in the end of the game but in the opening the handicap stones are just a burden and does not really make it play random. -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Anchor Player
Yes, in Chinese rules you need to compensate white for the extra area you gain from the actual stones. The handicap is only meant to be the extra strength/stability. To be slightly more specific, the extra compensation is specific to area scoring rule systems. In a game with only two passes, and black passes first, area scoring matches other systems. In a game with N handicap, however, an area scoring will give an extra (N-1) points to black. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
My reading of http://homepages.cwi.nl/~tromp/go.html is that komi is only supported among equal players, as agreed upon. However, this goes against the practice in Chinese and Ing rulesets, where white does recieve a komi to counterbalance the scoring of the additional stones placed by black. One rationale for this would be that it keeps the score comparable to the Japanese rulesets, where stones on the board do not count as territory. The server would have to make and enforce some particular rule. Regarding handicap stones, I'd suggest free placement, for consistency with Chinese scoring. Terry McIntyre UNIX for hire software development / systems administration / security - Original Message From: Christian Nilsson [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Friday, December 22, 2006 12:46:40 PM Subject: Re: [computer-go] Anchor Player Yes, in chinese rules you need to compensate white for the extra area you gain from the actual stones. The handicap is only meant to be the extra strength/stability. One can of course ignore this for the server. I just wanted to make sure all programs use the same rules. I don't know what the tromp-taylor rules says about it. On 12/22/06, Don Dailey [EMAIL PROTECTED] wrote: I'm trying to figure this out. If you get a 9 stone handicap, you have to give back those 9 stones? So a 9 stone handicap is not quite as much as it seems although it's still pretty good. - Don On Fri, 2006-12-22 at 21:24 +0100, Christian Nilsson wrote: There's also the small issue of the compensation given to white because of the extra black stones on the board. Setting a modified komi would break (MC-)programs with an internal rule for it. Not setting it would break those who does not use that rule. How is this compensation handled by the various programs on cgos, if at all? Check http://www.britgo.org/rules/compare.html#comp if you don't know what I'm talking about.. //Christian __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
So really, what I want to be able to do is: 1. Use the ELO rating system. 2. Determine how many ELO points 1 stone handicap is worth. 3. 2 stones are worth 4. 3 stones are worth, etc. When two players are matched, the server gives the handicap that most closely matches them up. For rating purposes when the game is over their ratings are adjusted to reflect the handicap. For example if 2 players a 400 ELO apart, they may get rated as if they are only 50 ELO different. It may even turn out that the weaker player has a higher expectancy of winning due to the handicap. We start with an initial estimation of what a stone is worth in ELO points, and let the server make small adjustments over time to make it match up with reality. Each extra stone of handicap is tracked separately because it's not likely the ELO gap is uniform. The adjustment to ELO is simple, we have an expectancy based on the ELO rating of both players (along with the handicap ELO compensation) and we have the actual results. We make a proportionate adjustment based on the difference between what we expected to happen and what actually did happen, which I think is exactly the same thing the ELO formula does! And it doesn't really matter which handicap system we use. I think I slightly prefer fixed placement handicaps, as I believe the weaker programs will benefit less from handicaps otherwise, which is the opposite of what we want. I don't plan to add compensation for the handicap stones. - Don On Fri, 2006-12-22 at 16:01 -0500, Don Dailey wrote: I'm glad you bring it up. From the same site, it appears there is no standard way of handling this. I will look to see what Tromp/Taylor says if anything. It would be nice if we could simple equate handicap with ELO points, I think it would be more accurate. We may find that 1 stone per kyu doesn't hold up forever. Then we just use ELO (converting in a straightforward way to kyu if we want this) and have a formula (or table) for compensating ELO.For instance we may determine that 400 ELO can be compensated by 4 stones and this effectively changes the ELO calculation. This seems more mathematically logical to me.Perhaps the server itself can converge on the right formula. - Don On Fri, 2006-12-22 at 21:46 +0100, Christian Nilsson wrote: Yes, in chinese rules you need to compensate white for the extra area you gain from the actual stones. The handicap is only meant to be the extra strength/stability. One can of course ignore this for the server. I just wanted to make sure all programs use the same rules. I don't know what the tromp-taylor rules says about it. On 12/22/06, Don Dailey [EMAIL PROTECTED] wrote: I'm trying to figure this out. If you get a 9 stone handicap, you have to give back those 9 stones? So a 9 stone handicap is not quite as much as it seems although it's still pretty good. - Don On Fri, 2006-12-22 at 21:24 +0100, Christian Nilsson wrote: There's also the small issue of the compensation given to white because of the extra black stones on the board. Setting a modified komi would break (MC-)programs with an internal rule for it. Not setting it would break those who does not use that rule. How is this compensation handled by the various programs on cgos, if at all? Check http://www.britgo.org/rules/compare.html#comp if you don't know what I'm talking about.. //Christian On 12/22/06, Don Dailey [EMAIL PROTECTED] wrote: Ok, Well I'm inclined to go with the majority which seems to have turned around from the last time I polled. Now the question: How to set it up? Here are the options: 1. Use GTP handicap commands to set up game. 2. Send the appropriate pass commands to get the initial setup to accommodate programs that have not implemented handicap. 3. Do both - send handicap to programs that can handle it, otherwise guide them through it by sending play commands with passes. In any case, I would make the game records (SGF) look correct, doing whatever that takes. - Don On Fri, 2006-12-22 at 20:48 +0100, Magnus Persson wrote: Quoting [EMAIL PROTECTED]: on 9x9 boards. To make a long story short, I didn't see any evidence that this algorithm is fundamentally disadvantaged in handicap games. In fact, I agree with Remi's view that it is particularly *well* suited to handicap games compared to territory based algorithms. When it finds itself behind, it goes for the swindle. Against an equal opponent, that's obnoxious and futile. Against a weaker foe, it's wise. I have to add to this. My
Re: [computer-go] Anchor Player
Le vendredi 22 décembre 2006 21:44, Don Dailey a écrit : [...] I still have a hard time believing that the system scales very well across a 9 kyu range. Handicap system works incredibly well, from very weak kyu to strong dan. Moreover, the problem of the black players are the same whatever his absolute strenght is! That's why some josekis are seen mainly in handicap games: they are simple, a little slow locally, but works perfectly with the others handicap stones, with no variation. As the handicap decrease, things are getting more complex, more variations are possible, global understanding is needed ... 9 handi means black is beginner and understand nearly nothing compared to white. Typical 9 handi problems are: - B corner is surrounded an die! - Black is so scared for his corners that it plays only pure defense move and makes no points. - W takes control of the center and kills a huge group. That's part of the magic of go. Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Increasing komi is much easier than placing stores, but a much weaker representation of how go games are actually played in the real world. cheers stuart On 12/15/06, Hideki Kato [EMAIL PROTECTED] wrote: Increasing KOMI is much easier than placing stones, right? Jacques Basaldúa‚³‚ñ [EMAIL PROTECTED]: I would like to take part in the 19x19 competition. I also prefer kyu rating to Elo, but I got the impression that you were relating kyu rating with handicap games (that is usually done by human players). I think handicap is a bad idea for computers. Handicap requires human intelligence to understand how the playing style must be changed. It completely ruins fuseki databases and may also make josekis that are good under equal play too slow. Of course, if you pretend to ruin fuseki database programs, its a good idea. But I think dan/pro level fuseki is not only legitimate, but probably the best possible fuseki and it can be played in ultrablitz which preserves computing time for later moves. The only drawback is 10 Mb of disk space. Any silly welcome video is heavier than that. I suggest, if handicap is implemented, it should be optional. Jacques. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Hi, Many win32 binary(such as Fritz) can run in linux with help of Wine(a free implementation of Windows on Unix) without noticeable performance loss. Best regards! - Original Message - From: Don Dailey [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Wednesday, December 13, 2006 2:47 AM Subject: [computer-go] Anchor Player If I set up a 19x19 server, we will need an Anchor player. Here is what I need from an Anchor player: 3. Linux binary - because it runs on the server itself. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
I run wine on my own computer, but it's not on the server computer and I believe it to be a resource hog.I want to keep it lean and simple on Dave Dyers server. - Don On Wed, 2006-12-13 at 16:46 +0800, Cai Qiang wrote: Hi, Many win32 binary(such as Fritz) can run in linux with help of Wine(a free implementation of Windows on Unix) without noticeable performance loss. Best regards! - Original Message - From: Don Dailey [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Wednesday, December 13, 2006 2:47 AM Subject: [computer-go] Anchor Player If I set up a 19x19 server, we will need an Anchor player. Here is what I need from an Anchor player: 3. Linux binary - because it runs on the server itself. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
Le Mercredi 13 Décembre 2006 05:56, Don Dailey a écrit : On Wed, 2006-12-13 at 04:48 +0100, [EMAIL PROTECTED] wrote: GnuGo is another possibility and has the advantage of being a well known quantity, but Gnugo fails to meet some of the criteria above such as being too deterministic and using heavy resources. But GnuGo uses a lot of memory, or perhaps that can be controlled but does that slow it down a lot? On my computer, gnugo at level 0 takes 13 Mbytes of memory and plays around 2 moves a second. I think it is ok no? I'm pretty sure gnugo is deterministic, need to check this out - maybe it would be ok to use gnugo if it isn't. Another way is to modify it to play more randomly during first few moves but I'm sure this weakens it and it wouldn't really represent gnugo. At least at the beginning gnugo is not deterministic whereas there are not so many variations. I think however it is random enough for an anchor. Of course gnugo for an anchor has also drawbacks, for example as the true version of gnugo would also play on the server. Sylvain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Anchor Player
I suggest you use anchorman. It will be weaker on 19x19, but so will the other programs. It lets you get set up quickly. David -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Don Dailey Sent: Tuesday, December 12, 2006 10:48 AM To: computer-go Subject: [computer-go] Anchor Player If I set up a 19x19 server, we will need an Anchor player. Here is what I need from an Anchor player: 1. Non-deterministic - should not play same game every time. 2. Consistent - plays at the same strength at a level that is not based on the power of the hardware. For instance AnchorMan is set to a fixed level that does not depend on time. Lazarus, however, players weaker when other jobs are running on the computer - something we don't want in an anchor. 3. Linux binary - because it runs on the server itself. 4. Low resource usage - I run AnchorMan on the server at a high nice level and it takes less than 1 second per move even if it isn't niced. If the Anchor runs on the server, it must be a good citizen. 5. Should play as strong as possible given the above constraints. If possible it should be in the upper 50-60 percentile - but it should not be significantly below median strength. It does not absolutely have to run on the server but it must be heavily available - pretty much 24 hours a day. It should be a non-changing entity - not something being constantly upgraded - although we could from time to time explicitly upgrade the Anchor player. It's better if the Anchor player is a known quantity on 9x9, then we could actually assign it the same rating and attempt to extrapolate, but we can do that anyway - not a big deal. The very best candidate may actually be AnchorMan, a program that may fit all the above criteria. It's an old fashioned Monte/Carlo program that plays about as well is at can and uses little memory given about 1 second per move - at least on 9x9. So it doesn't use much resources. At 19x19 AnchorMan would be weaker. At this boardsize, AnchorMan would benefit greatly from increased time control but then I'm starting to get away from constraint 4 - low resource usage - unless it was run remotely. GnuGo is another possibility and has the advantage of being a well known quantity, but Gnugo fails to meet some of the criteria above such as being too deterministic and using heavy resources. If someone wants to host an Anchor player remotely or has a resource friendly candidate that meets the above criteria, let me know. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Anchor Player
I suggest you use anchorman. It will be weaker on 19x19, but so will the other programs. It depends on the programs. Gnugo or Aya scale very well on 19x19. Then anchorMan would be far too weak for Aya and gnugo, and certainly other programs. But we can try some experiments, and perhaps change the anchor? Sylvain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/