Re: [computer-go] Re: Dynamic komi in commercial programs
I think the most promising field for using dynamic komi is in low handicap play. Because the main strategic principle for w in low handicap play is patience, which is easily and naturally modeled by a declining komi. For high handicaps the main strategic principle is light play, which is an elusive concept even for humans, and very difficult to even begin to implement. Light play is not the same as bean scattering. Maybe it will be possible to give greater weight to sparse shape patterns in high handicap go, but who knows? Stefan - Original Message - From: terry mcintyre To: computer-go Sent: Wednesday, July 15, 2009 12:29 AM Subject: Re: [computer-go] Re: Dynamic komi in commercial programs Maybe we should go back to the question which dynamic komi is an attempt to solve: how to obtain better discrimination when every move seems to be clustered near I am so freaking dead or I am so far ahead, as the case may be. Terry McIntyre terrymcint...@yahoo.com “We hang the petty thieves and appoint the great ones to public office.” -- Aesop -- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
On Tue, Jul 14, 2009 at 4:07 AM, Ingo Althöfer 3-hirn-ver...@gmx.dewrote: Darren Cook wrote: Ingo's suggestion (of two buttons to increment/decrement komi by one point) was to make it easy for strong humans to test out the idea for us. Don Dailey wrote: There is no question that if you provide a button to push, all kinds of positions will appear where this idea works. Providing a button is not nearly the same as providing an actual working algorithm that you can prove is superior. Right. Especially it may turn out that dynamic komi works well as a tool in computer-aided analysis. To give an example, the very new version 12.013 of Many Faces of Go has a feature: the program does not only compute its best move but also shows percentages for alternative popular moves (name coined by David Fotland). Here is a screenshot http://www.althofer.de/image-fotland.jpg or in context http://www.althofer.de/k-best-visualisations.html Of course, this feature not at all improves the playing strength of autonomous Many Faces. But it is a very hepful tool for computer- aided analysis of positions. The experience from almost two decades of chess programs on the pc: when you give users such tools to play with you will earlier or later get very helpful feedback. But this to me is just a random feature - it seems like many other more useful features would be much higher priority than a pseudo komi feature. - Don Ingo. -- Neu: GMX Doppel-FLAT mit Internet-Flatrate + Telefon-Flatrate für nur 19,99 Euro/mtl.!* http://portal.gmx.net/de/go/dsl02 ___ 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] Re: Dynamic komi in commercial programs
- Original Message - From: Dave Dyer dd...@real-me.net To: computer-go computer-go@computer-go.org Sent: Saturday, July 11, 2009 10:54 PM Subject: [computer-go] Re: Dynamic komi in commercial programs If you are in a lost position, good play is play that maximizes the probability of a turnaround, which is quite different depending on how far behind you are, and for what reason. If the status of all the major groups is solid, then concentrating on tactics which can gain a few points reliably might be the right thing. On the other hand, if the status of some groups is less than immutable, then focusing on changing their status favorably might be correct. It's hard to see how shifting Komi will influence the style of play in this direction. Dynamic komi in a sense means that the bot is deluding itself on purpose. Obviously this is dangerous medicine, a kind of magic mushroom. So what could be worse than a deluded bot? I say, letting a monkey play could be worse. And monkeys' play is what you get from an mc bot when all possible moves come back with indistinguishably wonderful or terrible win rates. Adding a wishful (or pessimistic) komi will distort reality, but will help create bigger win rate differences between moves. It should be possible to assign costs to both dynamic komi and to insufficiently spreading win rates between moves. My hypothesis is, that with unstable groups on the board, the win-rate spread will tend to be larger then with stable groups. So with proper balancing, the bot should refuse to take dynamic komi when he sees a chance to win outright. Stefan ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
Maybe we should go back to the question which dynamic komi is an attempt to solve: how to obtain better discrimination when every move seems to be clustered near I am so freaking dead or I am so far ahead, as the case may be. Terry McIntyre terrymcint...@yahoo.com “We hang the petty thieves and appoint the great ones to public office.” -- Aesop From: Stefan Kaitschick stefan.kaitsch...@hamburg.de To: computer-go computer-go@computer-go.org Sent: Tuesday, July 14, 2009 3:17:38 PM Subject: Re: [computer-go] Re: Dynamic komi in commercial programs Dynamic komi in a sense means that the bot is deluding itself on purpose. Obviously this is dangerous medicine, a kind of magic mushroom. So what could be worse than a deluded bot? I say, letting a monkey play could be worse. And monkeys' play is what you get from an mc bot when all possible moves come back with indistinguishably wonderful or terrible win rates. Adding a wishful (or pessimistic) komi will distort reality, but will help create bigger win rate differences between moves. It should be possible to assign costs to both dynamic komi and to insufficiently spreading win rates between moves. My hypothesis is, that with unstable groups on the board, the win-rate spread will tend to be larger then with stable groups. So with proper balancing, the bot should refuse to take dynamic komi when he sees a chance to win outright. Stefan ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
So changing the komi doesn't actually improve your confidence interval. If (as Darren said) the win percentage is a crude estimate of the final score, then changing komi would do nothing to change the results one got (and at extremes biases it badly). Moving the ratios closer to 50/50 (by whatever means) at the high end changes the variance of the data, and in a world where there's a 1:1 correspondence between score and win ratio does nothing to change one's confidence that the highest ranked node should be. Of note there is that the goal, of any method chosen, is to make the ranking of individual moves as accurate as possible. It can do that by either increasing the number of simulations or by increasing the granularity of the metric. This second point seems more like a testbed for that sigmoid function paper than for dynamic komi, but that's just my guess. Of course what would be most preferable would be lots of data. Arguing with guesses instead of data is silly. ~ 2009/7/14 terry mcintyre terrymcint...@yahoo.com: Maybe we should go back to the question which dynamic komi is an attempt to solve: how to obtain better discrimination when every move seems to be clustered near I am so freaking dead or I am so far ahead, as the case may be. Terry McIntyre terrymcint...@yahoo.com “We hang the petty thieves and appoint the great ones to public office.” -- Aesop From: Stefan Kaitschick stefan.kaitsch...@hamburg.de To: computer-go computer-go@computer-go.org Sent: Tuesday, July 14, 2009 3:17:38 PM Subject: Re: [computer-go] Re: Dynamic komi in commercial programs Dynamic komi in a sense means that the bot is deluding itself on purpose. Obviously this is dangerous medicine, a kind of magic mushroom. So what could be worse than a deluded bot? I say, letting a monkey play could be worse. And monkeys' play is what you get from an mc bot when all possible moves come back with indistinguishably wonderful or terrible win rates. Adding a wishful (or pessimistic) komi will distort reality, but will help create bigger win rate differences between moves. It should be possible to assign costs to both dynamic komi and to insufficiently spreading win rates between moves. My hypothesis is, that with unstable groups on the board, the win-rate spread will tend to be larger then with stable groups. So with proper balancing, the bot should refuse to take dynamic komi when he sees a chance to win outright. Stefan ___ 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] Re: Dynamic komi in commercial programs
I also tried dynamic komi with Valkyria a long time ago. It failed. I did not waste much time on it. Anyway here are my opinions and intuitions about it. As usual I am open to been proved being wrong with some empirical evidence along with a nice algorithm that I can steal and add to Valkyria. :-) As already mentioned, one needs to set the dynamic komi to some kind of magic value, which IMO requires so much insight about the position, it means that the program should *know* the best way of playing anyway without the help of the dynamic komi. Personally I am absolutely happy with how Valkyria plays in handicap games. It gives large handicaps against weaker players on KGS and wins. The only problem is to avoid it resigning early because the situation is hopeless. With handicap on small boards it is so strong that the problem just do not seem to exist anymore. Sure it plays ugly moves in handicap games, but this is not different from even games where it often plays a little bit too creative to my taste. The problem with MCTS is that they are weak in evaluation. I am pretty sure that this fixation on dynamic Komi is confounded with the fact that most programs has quite light playouts, and programs with heavy playouts still has big holes in the knowledge that leads to delusional evaluations. I guess the reason that people think dynamic komi is important is that these bad evaluations can always in principle be repaired in *hindsight*. But repairing something in hindsight is useless. If I fiddle with komi until the program plays a move I like it is not the program that selects the move anymore. As the programmer I cannot add a human expert to the code. This is most certainly true for 19x19. The correct solution to bad plays is to make the program *stronger*. I am open to opponent modeling such as make the playouts of black in handicap games weaker. But in this case I think real gain if any would come from making the statistics more sensitive to the qualitative difference in available moves, rather than actually modeling the opponent, by bringing the win rates closer to 50%. Although I think it would be really hard to degrade the black moves in the playouts in a realistic way. -Magnus ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
Is it possible to design metrics for complexity of positions? An opponent model could make use of that information; there are positions which some players will totally fail to grok. Double-digit kyu players are weak on life-and-death, ko, and seki. Some otherwise strong programs will fail to read seki properly. As players move up in rank, they read life-and-death at an earlier point. Human players, with practice, discover the weaknesses of their particular opponents. Terry McIntyre terrymcint...@yahoo.com “We hang the petty thieves and appoint the great ones to public office.” -- Aesop From: Magnus Persson magnus.pers...@phmp.se I am open to opponent modeling such as make the playouts of black in handicap games weaker. But in this case I think real gain if any would come from making the statistics more sensitive to the qualitative difference in available moves, rather than actually modeling the opponent, by bringing the win rates closer to 50%. Although I think it would be really hard to degrade the black moves in the playouts in a realistic way. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
Ingo, are you sure you already want to bet on one particular technique? :) I don't believe a score optimisation algorithm like UCT works all that well when behind. I am pretty sure that human players do *not* choose between the top three moves if their values are 40%, 39% and 38%. They will start looking for other moves. I am also not sure the dynamic komi will help, if all it ends up doing is shifting the percentages. Criteria for moves for humans when playing behind may look something like this (subjective assessment): 1. Am I certain of the score after the move, and does it lead to a loss? Discard it, there is no point - unless it is a waiting sequence to prepare for another move, or a move that answers a sente play. 2. Does the move create uncertainty in the score? One does not play for certainty when behind, but for confusion. If I cannot see the result, my opponent probably can't either. 3. Is it a meltdown move (i.e. if it goes wrong, I will definitely lose the game)? If so, I might not want to play it unless I am very far behind ( 10 points, and in the middle game) I wonder if anybody has looked at alternating the evaluation function when behind? This is, of course, very difficult without providing a point of attack or discontinuity in its playing style. Perhaps moves with a higher standard deviation could be chosen once in a while? Whatever the case may be, I agree that things definitely need to be tried out, against strong players, perhaps on KGS. I gave Zen, an excellent program, 2 stones yesterday night, and sure enough it melted down spectacularly once it was behind :) Christian p.s. I believe handicap games need to be treated differently from situations where one is behind in even games - in the former, one can wait for the opponent's mistakes; in the second, one needs to be proactive. Ingo Althöfer wrote: Don Dailey wrote: I think we should open up to other ideas, not just dynamic komi modification. In fact that has not proved to be a very fruitful technique and I don't understand the fascination with it. I was not clear enough in the original posting. My main point is the following: Currently the programmers are developing program, and customers are playing with them. But we should try to bring creative customers inside the process of development. Dynamic komi might be one good early try for this approach. You write that [DyKo] has not proved to be a very fruitful technique but you are right only concerning the rather narrow community of programmers. (Some) creative customers have lots of time to test strange things and settings and are willing to do so, provided the programs are sufficiently test-friendly. Concerning dynamic komi I would bet 1:1 that their findings might lead to a breakthrough. Ingo. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
Hi, I would like to know what exact experiments with virtual komi have been made and why thay failed. To me, this idea seems very natural, as it encodes the confidence of the stronger player that the weaker one will eventually make more mistakes on his own. You don't need to catch up a fourty-point handicap at once and try to kill all - instead you just overplay a little in order to catch up slowly but steadily. If you're behind by 5 points after move 100 against a player who is five stones weaker than you, you can almost consider it a sure win. If you're behind by the same amount, but when the last endgame moves are being played, it's a safe loss. This all is encoded very naturally by a decreasing virtual komi. So why exactly shouldn't it work? Cheers, Benjamin ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
On Sun, Jul 12, 2009 at 8:07 AM, Benjamin Teuber benjamin.teu...@web.dewrote: Hi, I would like to know what exact experiments with virtual komi have been made and why thay failed. To me, this idea seems very natural, as it encodes the confidence of the stronger player that the weaker one will eventually make more mistakes on his own. You don't need to catch up a fourty-point handicap at once and try to kill all - instead you just overplay a little in order to catch up slowly but steadily. You just hit the nail on the head. Dynamic komi does not encourage a program to overplay the position. Since you are starting from a losing position you HAVE to overplay a bit. You have to attack when it is futile. Dynamic komi just makes the program happy with less.That is NOT a good algorithm for winning against fallible opponents when you are behind.It' s NOT a natural algorithm and I don't believe it's what humans do either. Dynamic komi doesn't tell the program that you should fight for something that you probably cannot win - which is what you have to do in handicap play. It just tells the program that it's ok not to fight and play as if everything is fine. What I'm suggesting is not to ignore the problem but to find some other technique that actually addresses the true problem. If you're behind by 5 points after move 100 against a player who is five stones weaker than you, you can almost consider it a sure win. If you're behind by the same amount, but when the last endgame moves are being played, it's a safe loss. This all is encoded very naturally by a decreasing virtual komi. So why exactly shouldn't it work? Cheers, Benjamin ___ 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] Re: Dynamic komi in commercial programs
Don Dailey wrote: You just hit the nail on the head. Dynamic komi does not encourage a program to overplay the position. Since you are starting from a losing position you HAVE to overplay a bit. You have to attack when it is futile. Dynamic komi just makes the program happy with less. That is NOT a good algorithm for winning against fallible opponents when you are behind. It's NOT a natural algorithm and I don't believe it's what humans do either. It is how I was taught to play when giving a handicap: don't overplay, let your opponent make their own mistakes. This is partly because handicap games are traditionally supposed to be teaching games: you're aiming to set a good example, not to win at all costs. But I've also found that avoiding conscious overplays in handicap games is a good winning strategy. -M- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
Benjamin Teuber wrote: I would like to know what exact experiments with virtual komi have been made and why thay failed. In particular, it would be interesting to know what board sizes people have tried it with. -M- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
You just hit the nail on the head. Dynamic komi does not encourage a program to overplay the position. Since you are starting from a losing position you HAVE to overplay a bit. You have to attack when it is futile. That depends on the komi - if you're behind by fourty points and set the virtual komi to 30, you play as if you'd be 10 behind, which would be agressive, but not kamikaze. This is exactly what people do, so I don't see your point. Benjamin ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
On Sun, Jul 12, 2009 at 8:59 AM, Benjamin Teuber benjamin.teu...@web.dewrote: You just hit the nail on the head. Dynamic komi does not encourage a program to overplay the position. Since you are starting from a losing position you HAVE to overplay a bit. You have to attack when it is futile. That depends on the komi - if you're behind by fourty points and set the virtual komi to 30, you play as if you'd be 10 behind, which would be agressive, but not kamikaze. This is exactly what people do, so I don't see your point. It's not up to me to prove anything. It's up to you. Several of us have tried variations of this idea of dynamic komi adjustment, which seems like a very good premise. This SHOULD help the play.But the fact of the matter is that none of us (so far) has made it work. If the observations do not fit the premise, at some point we should actually scrutinize the premise - even if the premise seems logical to us. I think the ones who still cling to this idea have not actually tried implementing it.Have you tried?If you have, why are still talking about it and not showing us something? - Don Benjamin ___ 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] Re: Dynamic komi in commercial programs
There are 3 commonly cited problems and 4 commonly proposed remedies. Problems: 1) Human games remain interesting, even after the winner is clear, because the players just naturally switch to playing for maximum territory. Wouldn't MCTS bots be more fun to play against if they did that too? 2) Sometimes a bot has a win by a small margin, but thinks it's a win by a big margin (because it is misreading a seki or whatever). Consequently, the bot neglects to defend the space that matters and loses after all. 3) For a big enough handicap, the bot plays random, ugly looking moves in the beginning. Can't that be improved? Remedies: a) Play for maximum territory sometimes. b) Fake the Komi sometimes. c) Unbalance the playout strength sometimes. d) Worry about more important things. The vagueness in the sometimes part doesn't help in deciding which remedy is best for which problem. Looking at the handicap problem alone, how can I pick the best remedy and justify my decision? Maybe I could take my engine at a reasonable time setting and experiment with all the remedies to try to find the highest handicap I can give Wally and still win 50% of the games. - Dave Hillis ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
Don, others, are there publications about this? If people have tried it out, are there any threads on this list that somebody remembers where results are posted? I have not been able to find any. It would be interesting to see. Christian 2009/7/12 Don Dailey dailey@gmail.com: On Sun, Jul 12, 2009 at 8:59 AM, Benjamin Teuber benjamin.teu...@web.de wrote: You just hit the nail on the head. Dynamic komi does not encourage a program to overplay the position. Since you are starting from a losing position you HAVE to overplay a bit. You have to attack when it is futile. That depends on the komi - if you're behind by fourty points and set the virtual komi to 30, you play as if you'd be 10 behind, which would be agressive, but not kamikaze. This is exactly what people do, so I don't see your point. It's not up to me to prove anything. It's up to you. Several of us have tried variations of this idea of dynamic komi adjustment, which seems like a very good premise. This SHOULD help the play. But the fact of the matter is that none of us (so far) has made it work. If the observations do not fit the premise, at some point we should actually scrutinize the premise - even if the premise seems logical to us. I think the ones who still cling to this idea have not actually tried implementing it. Have you tried? If you have, why are still talking about it and not showing us something? - Don Benjamin ___ 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] Re: Dynamic komi in commercial programs
e) use a knowledge system that knows what good moves look to prune or bias the moves when way ahead or way behind. This is what many Faces does. David From: computer-go-boun...@computer-go.org [mailto:computer-go-boun...@computer-go.org] On Behalf Of dhillism...@netscape.net Sent: Sunday, July 12, 2009 9:54 AM To: computer-go@computer-go.org Subject: Re: [computer-go] Re: Dynamic komi in commercial programs There are 3 commonly cited problems and 4 commonly proposed remedies. Problems: 1) Human games remain interesting, even after the winner is clear, because the players just naturally switch to playing for maximum territory. Wouldn't MCTS bots be more fun to play against if they did that too? 2) Sometimes a bot has a win by a small margin, but thinks it's a win by a big margin (because it is misreading a seki or whatever). Consequently, the bot neglects to defend the space that matters and loses after all. 3) For a big enough handicap, the bot plays random, ugly looking moves in the beginning. Can't that be improved? Remedies: a) Play for maximum territory sometimes. b) Fake the Komi sometimes. c) Unbalance the playout strength sometimes. d) Worry about more important things. The vagueness in the sometimes part doesn't help in deciding which remedy is best for which problem. Looking at the handicap problem alone, how can I pick the best remedy and justify my decision? Maybe I could take my engine at a reasonable time setting and experiment with all the remedies to try to find the highest handicap I can give Wally and still win 50% of the games. - Dave Hillis _ A Good Credit Score is 700 or Above. See http://pr.atwola.com/promoclk/100126575x1222377098x1201454399/aol?redir=htt p://www.freecreditreport.com/pm/default.aspx?sc=668072%26hmpgID=62%26bcd=Jul ystepsfooterNO62 yours in just 2 easy steps! ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
It's not up to me to prove anything. It's up to you. You entered a discussion in which you gave arguments (that I believe are nonsense) against this method, which I just meant to counter. But I don't want to prove anything (well I might want, but I know I cannot). I'm really just curious about this good-sounding idea and hoped somebody might be able to give details of its failure. Cheers, Benjamin ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
2009/7/12 David Fotland fotl...@smart-games.com e) use a knowledge system that knows what good moves look to prune or bias the moves when way ahead or way behind. This is what many Faces does. This is what I believe to be the most reasonable approach. - Don David *From:* computer-go-boun...@computer-go.org [mailto: computer-go-boun...@computer-go.org] *On Behalf Of * dhillism...@netscape.net *Sent:* Sunday, July 12, 2009 9:54 AM *To:* computer-go@computer-go.org *Subject:* Re: [computer-go] Re: Dynamic komi in commercial programs There are 3 commonly cited problems and 4 commonly proposed remedies. Problems: 1) Human games remain interesting, even after the winner is clear, because the players just naturally switch to playing for maximum territory. Wouldn't MCTS bots be more fun to play against if they did that too? 2) Sometimes a bot has a win by a small margin, but thinks it's a win by a big margin (because it is misreading a seki or whatever). Consequently, the bot neglects to defend the space that matters and loses after all. 3) For a big enough handicap, the bot plays random, ugly looking moves in the beginning. Can't that be improved? Remedies: a) Play for maximum territory sometimes. b) Fake the Komi sometimes. c) Unbalance the playout strength sometimes. d) Worry about more important things. The vagueness in the sometimes part doesn't help in deciding which remedy is best for which problem. Looking at the handicap problem alone, how can I pick the best remedy and justify my decision? Maybe I could take my engine at a reasonable time setting and experiment with all the remedies to try to find the highest handicap I can give Wally and still win 50% of the games. - Dave Hillis -- *A Good Credit Score is 700 or Above. See yours in just 2 easy steps!http://pr.atwola.com/promoclk/100126575x1222377098x1201454399/aol?redir=http://www.freecreditreport.com/pm/default.aspx?sc=668072%26hmpgID=62%26bcd=JulystepsfooterNO62 * ___ 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] Re: Dynamic komi in commercial programs
On Sun, Jul 12, 2009 at 1:22 PM, Christian Nentwich christian.nentw...@gmail.com wrote: Don, others, are there publications about this? If people have tried it out, are there any threads on this list that somebody remembers where results are posted? I have not been able to find any. It would be interesting to see. I think I just mentioned that there is probably not much on this except in the archive.And even then it's probably not very well documented. I did try this myself but I don't have any data to show what I did.What I remember is that it's incredibly tricky - how do you actually know when and how much to adjust? If the score starts getting really low or really high, do you restart the search with a new komi?If you restart then you have wasted effort. I tried 2 different thing. One of them involved using the total points won in some kind of hybrid approach and the other involved changing the komi during the game. Using JUST the total points won is a drastic weakening of the program and it's surprising how much. I tried factoring in a percentage of total points won and other things.After some time I gave up - it seemed like I was taking something that worked well and trying to make it better by factoring in something that sucked. It was like trying to make it play better by putting something in on purpose that I knew makes it play worse. The dynamic komi adjustment, from my recollection was more promising, but still played worse. The only way to make this work is if you know in advance what kind of position you really have.If you KNOW that you can pick off a small group without risk, then it probably would work just fine.But just increasing komi for no reason except that you are winning is not good enough. For instance if you KNOW there is a seki issue, then you should probably do it.But just doing it because there MIGHT be a seki issue every 50 games that actually matters is not good enough. You could call this a chicken and egg problem.You can of course easily construct positions that will illustrate how wonderful the idea is, and it will probably work great in those positions.Seki positions are always given as to why this will help - but not every game has a game critical seki. But I'm pretty convinced you cannot generalize the idea. You would have to do some kind of pre-analsysis to figure out what needs to be done, and by then you may already know what to do anyway and you have a more convential program. - Don Christian 2009/7/12 Don Dailey dailey@gmail.com: On Sun, Jul 12, 2009 at 8:59 AM, Benjamin Teuber benjamin.teu...@web.de wrote: You just hit the nail on the head. Dynamic komi does not encourage a program to overplay the position. Since you are starting from a losing position you HAVE to overplay a bit. You have to attack when it is futile. That depends on the komi - if you're behind by fourty points and set the virtual komi to 30, you play as if you'd be 10 behind, which would be agressive, but not kamikaze. This is exactly what people do, so I don't see your point. It's not up to me to prove anything. It's up to you. Several of us have tried variations of this idea of dynamic komi adjustment, which seems like a very good premise. This SHOULD help the play.But the fact of the matter is that none of us (so far) has made it work. If the observations do not fit the premise, at some point we should actually scrutinize the premise - even if the premise seems logical to us. I think the ones who still cling to this idea have not actually tried implementing it.Have you tried?If you have, why are still talking about it and not showing us something? - Don Benjamin ___ 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] Re: Dynamic komi in commercial programs
On Sun, Jul 12, 2009 at 3:08 PM, Benjamin Teuber benjamin.teu...@web.dewrote: It's not up to me to prove anything. It's up to you. You entered a discussion in which you gave arguments (that I believe are nonsense) ... but at least fits the observation that this method does not work. ... against this method, which I just meant to counter. Why don't you counter it with an argument that fits the actual observation? But I don't want to prove anything (well I might want, but I know I cannot). I'm really just curious about this good-sounding idea and hoped somebody might be able to give details of its failure. There are some things in the archive on this - the idea is literally years old. But I doubt you will any papers on this since it's a failure. Paper authors tend to spend more time writing papers on things that work - unless they have some really surprising or interesting result to report. - Don Cheers, Benjamin ___ 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] Re: Dynamic komi in commercial programs
Don Dailey wrote: I did try this myself but I don't have any data to show what I did.What I remember is that it's incredibly tricky - how do you actually know when and how much to adjust? If the score starts getting really low or really high, do you restart the search with a new komi?If you restart then you have wasted effort. [...] The dynamic komi adjustment, from my recollection was more promising, but still played worse. So what board sizes were you working with? -M- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
On Sun, Jul 12, 2009 at 7:12 PM, Matthew Woodcraft matt...@woodcraft.me.ukwrote: Don Dailey wrote: I did try this myself but I don't have any data to show what I did. What I remember is that it's incredibly tricky - how do you actually know when and how much to adjust? If the score starts getting really low or really high, do you restart the search with a new komi?If you restart then you have wasted effort. [...] The dynamic komi adjustment, from my recollection was more promising, but still played worse. So what board sizes were you working with? I only worked with boardsize 9. - Don -M- ___ 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] Re: Dynamic komi in commercial programs
I would like to know what exact experiments with virtual komi have been made and why thay failed. ... I'm only aware of Don's experiment [1], which he admits he doesn't have any details for and only remembers: I did a bunch of experiments and ALWAYS got a reduced wins when I faked the komi. On the other side we have some experiments by Kato-san [2] (where he reports a 100 ELO improvement over GnuGo, but only from a few tens of game) and a subjective experiment by Okasaki-san where he reported Mogo played clearly stronger on KGS [3]. My own experiments are even more subjective and small-scale, and in the context of 9x9 endgames, not 19x19 handicap openings. However they were enough to make me think the technique is viable, but that if you don't adjust the komi down so the winning rate is near 50% it is wasted effort (*), and so you need to replay the same move over and over with different komi until you zero in on that point. *: I.e. the program still plays weak moves if you've only adjusted komi to go from 80% to 65%, or from 25% to 35%. kill all - instead you just overplay a little in order to catch up slowly but steadily. You just hit the nail on the head. Dynamic komi does not encourage a program to overplay the position. Since you are starting from a losing position you HAVE to overplay a bit. You have to attack when it is futile. If the handicap is correct then you don't really need to overplay. As the stronger player you might guide the game towards more complex positions to encourage more mistakes, but mainly you are just sitting around waiting for those inevitable mistakes. But, the real point of adjusting komi is it is an easy to understand way to overcome MCTS's problem when seeing all moves as winning/losing, and choosing effectively randomly instead of falling back on an opponent model as a human would do. Ingo's suggestion (of two buttons to increment/decrement komi by one point) was to make it easy for strong humans to test out the idea for us. Darren [1]: http://computer-go.org/pipermail/computer-go/2008-August/015870.html [2]: http://computer-go.org/pipermail/computer-go/2008-February/014283.html [3]: http://computer-go.org/pipermail/computer-go/2008-August/015877.html ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
... 3) For a big enough handicap, the bot plays random, ugly looking moves in the beginning. Can't that be improved? Remedies: ... Another remedy is to have some handicap opening books, just to help get the MCTS programs get a bit further along before they start their all moves are created equal low-quality play. Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/gobet/ (Shodan Go Bet - who will win?) http://dcook.org/mlsn/ (Multilingual open source semantic network) http://dcook.org/work/ (About me and my work) http://dcook.org/blogs.html (My blogs and articles) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Dynamic komi in commercial programs
On Sun, Jul 12, 2009 at 8:10 PM, Darren Cook dar...@dcook.org wrote: I would like to know what exact experiments with virtual komi have been made and why thay failed. ... I'm only aware of Don's experiment [1], which he admits he doesn't have any details for and only remembers: I did a bunch of experiments and ALWAYS got a reduced wins when I faked the komi. On the other side we have some experiments by Kato-san [2] (where he reports a 100 ELO improvement over GnuGo, but only from a few tens of game) and a subjective experiment by Okasaki-san where he reported Mogo played clearly stronger on KGS [3]. My own experiments are even more subjective and small-scale, and in the context of 9x9 endgames, not 19x19 handicap openings. However they were enough to make me think the technique is viable, but that if you don't adjust the komi down so the winning rate is near 50% it is wasted effort (*), and so you need to replay the same move over and over with different komi until you zero in on that point. *: I.e. the program still plays weak moves if you've only adjusted komi to go from 80% to 65%, or from 25% to 35%. kill all - instead you just overplay a little in order to catch up slowly but steadily. You just hit the nail on the head. Dynamic komi does not encourage a program to overplay the position. Since you are starting from a losing position you HAVE to overplay a bit. You have to attack when it is futile. If the handicap is correct then you don't really need to overplay. As the stronger player you might guide the game towards more complex positions to encourage more mistakes, but mainly you are just sitting around waiting for those inevitable mistakes. But, the real point of adjusting komi is it is an easy to understand way to overcome MCTS's problem when seeing all moves as winning/losing, and choosing effectively randomly instead of falling back on an opponent model as a human would do. Ingo's suggestion (of two buttons to increment/decrement komi by one point) was to make it easy for strong humans to test out the idea for us. There is no question that if you provide a button to push, all kinds of positions will appear where this idea works. Providing a button is not nearly the same as providing an actual working algorithm that you can prove is superior. So if you can do this in a verifiable way I'll be interested. - Don Darren [1]: http://computer-go.org/pipermail/computer-go/2008-August/015870.html [2]: http://computer-go.org/pipermail/computer-go/2008-February/014283.html [3]: http://computer-go.org/pipermail/computer-go/2008-August/015877.html ___ 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] Re: Dynamic komi in commercial programs
That's an interesting idea - factoring in knowledge about the variability of a position. Certain parts of the board are going to be stable with alternating play - you attack, I defend, the position remains stable. Other parts of the board are less well-defined. On the 9x9 board, conflicts easily spill over; everything is inter-connected until the end-game. On the 19x19 board, it's possible to establish stable groups, and the action usually happens on the borders. Factor in seki, ko fights, etc. Simply varying komi as some pre-ordained function of move number is unlikely to have enough granularity to cope with the structure of a particular game. Terry McIntyre terrymcint...@yahoo.com “We hang the petty thieves and appoint the great ones to public office.” -- Aesop From: Don Dailey dailey@gmail.com To: computer-go computer-go@computer-go.org Sent: Sunday, July 12, 2009 3:40:10 PM Subject: Re: [computer-go] Re: Dynamic komi in commercial programs On Sun, Jul 12, 2009 at 1:22 PM, Christian Nentwich christian.nentw...@gmail.com wrote: Don, others, are there publications about this? If people have tried it out, are there any threads on this list that somebody remembers where results are posted? I have not been able to find any. It would be interesting to see. I think I just mentioned that there is probably not much on this except in the archive.And even then it's probably not very well documented. I did try this myself but I don't have any data to show what I did.What I remember is that it's incredibly tricky - how do you actually know when and how much to adjust? If the score starts getting really low or really high, do you restart the search with a new komi?If you restart then you have wasted effort. I tried 2 different thing. One of them involved using the total points won in some kind of hybrid approach and the other involved changing the komi during the game. Using JUST the total points won is a drastic weakening of the program and it's surprising how much. I tried factoring in a percentage of total points won and other things.After some time I gave up - it seemed like I was taking something that worked well and trying to make it better by factoring in something that sucked. It was like trying to make it play better by putting something in on purpose that I knew makes it play worse. The dynamic komi adjustment, from my recollection was more promising, but still played worse. The only way to make this work is if you know in advance what kind of position you really have.If you KNOW that you can pick off a small group without risk, then it probably would work just fine.But just increasing komi for no reason except that you are winning is not good enough. For instance if you KNOW there is a seki issue, then you should probably do it. But just doing it because there MIGHT be a seki issue every 50 games that actually matters is not good enough. You could call this a chicken and egg problem.You can of course easily construct positions that will illustrate how wonderful the idea is, and it will probably work great in those positions.Seki positions are always given as to why this will help - but not every game has a game critical seki. But I'm pretty convinced you cannot generalize the idea. You would have to do some kind of pre-analsysis to figure out what needs to be done, and by then you may already know what to do anyway and you have a more convential program. - Don Christian 2009/7/12 Don Dailey dailey@gmail.com: On Sun, Jul 12, 2009 at 8:59 AM, Benjamin Teuber benjamin.teu...@web.de wrote: You just hit the nail on the head. Dynamic komi does not encourage a program to overplay the position. Since you are starting from a losing position you HAVE to overplay a bit. You have to attack when it is futile. That depends on the komi - if you're behind by fourty points and set the virtual komi to 30, you play as if you'd be 10 behind, which would be agressive, but not kamikaze. This is exactly what people do, so I don't see your point. It's not up to me to prove anything. It's up to you. Several of us have tried variations of this idea of dynamic komi adjustment, which seems like a very good premise. This SHOULD help the play.But the fact of the matter is that none of us (so far) has made it work. If the observations do not fit the premise, at some point we should actually scrutinize the premise - even if the premise seems logical to us. I think the ones who still cling to this idea have not actually tried implementing it.Have you tried?If you have, why are still talking about it and not showing us something? - Don Benjamin ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer
Re: [computer-go] Re: Dynamic komi in commercial programs
On Sat, Jul 11, 2009 at 4:54 PM, Dave Dyer dd...@real-me.net wrote: If you are in a lost position, good play is play that maximizes the probability of a turnaround, which is quite different depending on how far behind you are, and for what reason. What maximizes the probability of a turnaround depends on your opponent more than anything else. I'm sure the best move by this definition will change according to who you are playing. If the status of all the major groups is solid, then concentrating on tactics which can gain a few points reliably might be the right thing. I think the best PRACTICAL definition (which can be formalized) is to play the move (or one of the moves) that maximizes the total points on the board.I think this is the natural human style, more or less. My real point is that whether a move is good or bad cannot be precisely defined if you are looking for a practical definition. If you use my theoretical definion, it can be precisely defined, but it may not be the best practical definition for winning real games against fallible opponents. - Don On the other hand, if the status of some groups is less than immutable, then focusing on changing their status favorably might be correct. It's hard to see how shifting Komi will influence the style of play in this direction. ___ 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] Re: Dynamic komi in commercial programs
i think that the rationale behind variable komi is intuitive: good players can handicap one another more effectively with komi than with handicap stones, because it's more fine-grained. this is likely what is leading to the idea that computers playing handicap games could use this to their advantage. there's something like an exact tradeoff between stones and komi, although i don't know the function and it'd be interesting to find. s. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
