Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
> Petr Baudis wrote: > >> MoGo can indeed play out some rather spectacular ko fights; > >> unfortunately, I couldn't find any quickly, so here is at least an > >> example of a shorter one. > I see you made the following comment in that game record, which seems > relevant to recent discussions here. > | mogo excels at reducing clear winning positions to close games they > | lose because of botched up tsumego > Is it mogo botching up the tsumego or its opponents? Do you have any > example game records for this? You won't find that in computer vs computer games, because "tricking" the strong programs requires some go skill and it only works if you wait long enough before you "solve" the position. But if you search KGS (LeelaBot, CrazyStone, CzechBot) for even games where the bot lost against a kyu players you will find many. All go more or less like that: A 4-6 kyu human is behind by 10-15 points in the midgame (at that stage the probability of winning is correlated with territory, so the MC bot is building fine.) He creates a 12-16 point worth nakade trick in a corner and does not solve it.The bot is happy, it thinks a bulk five is alive or something like that. Perhaps the human sacrificed another 15 points somewhere to create the trick so he should be dead lost. But, he only has to play on, reduce, etc. As the endgame approaches, the MC bot allows the reduction only until the territorial balance would change the winner. The player is happy, he turned a 25 points loss into a 1.5 point loss (assumed by the program) and has a 12 point surprise. At the end, when the whole board is decided, the player kills the bot's group and the bot turns a sure win into a sure loss and resigns. Because the trick can only be played by similar strength players (much weaker players can't build something like that, much stronger don't need it) it affects the rating of the bots. I guess CrazyStone could be near KGS 1dan with that solved. It is 2k now. But, of course, the solution may not come at the price of making the program weaker. That is the difficult part. Jacques. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Floating komi
Christoph Birk wrote: > On Mar 5, 2008, at 11:58 AM, Don Dailey wrote: >>> Don Dailey wrote: not assuming that MC plays the best move. The problem isn't the >> assumptions I am making, but the assumptions others are making, that >> it's NOT playing the best move.You want to apply a fix to all >> positions without really knowing which positions are a problem. > > One last time: Nobody suggested a one fix for all positions/problems. > The "floating komi" was suggested to guide the UCT search along > certain lines of play during specific (close!) endgame positions. When I said all positions I meant all games.You expect to apply this to all winning and losing positions in every game, not just specific ones. - Don > > Christoph > > ___ > 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: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
> > > You won't find that in computer vs computer games, because "tricking" the > strong programs requires some go skill and it only works if you wait long > enough before you "solve" the position. But if you search KGS (LeelaBot, > CrazyStone, CzechBot) for even games where the bot lost against a kyu > players you will find many. All go more or less like that: > > A 4-6 kyu human is behind by 10-15 points in the midgame (at that > stage the > probability of winning is correlated with territory, so the MC bot is > building fine.) He creates a 12-16 point worth nakade trick in a corner > and does not solve it.The bot is happy, it thinks a bulk five is alive or > something like that. Perhaps the human sacrificed another 15 points > somewhere to create the trick so he should be dead lost. But, he only > has to play on, reduce, etc. As the endgame approaches, the MC bot > allows the reduction only until the territorial balance would change the > winner. The player is happy, he turned a 25 points loss into a 1.5 point > loss (assumed by the program) and has a 12 point surprise. > At the end, when the whole board is decided, the player kills > the bot's group and the bot turns a sure win into a sure loss and > resigns. > > Because the trick can only be played by similar strength players (much > weaker players can't build something like that, much stronger don't > need it) > it affects the rating of the bots. I guess CrazyStone could be near > KGS 1dan > with that solved. It is 2k now. But, of course, the solution may not > come at > the price of making the program weaker. That is the difficult part. I want to make sure I understand the nakade problem, please correct me if I am wrong: My understanding of this is that many program do not allow self-atari moves in the play-outs because in general the overwhelming majority are stupid moves. Is that what is causing the nakade problem? And if you start including self-atari you weaken the program in general? And can I assume the tree portion is also inhibited from seeing this due to a combination of factors such as heuristics to delay exploring "ugly" moves as well as the weakness of the play-outs in this regard (which would cause the tree to not be inclined to get close enough to the issue to understand it properly?) - Don > > > 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] Floating komi
steve uurtamo wrote: > why doesn't someone simply try this and post the results, > if they think that it would help? > > If someone would do this scientifically I would be all for it. But I am a bit of a pessimist about this. I can easily imagine someone trying it at reporting good results when the program is actually weaker. They would say, "yeah, I tried it and it worked pretty good!"or "Yeah, I looked at some games and it was playing much better." I would be satisfied if someone implemented it, reported a 500 game self-test sample and concluded that it didn't hurt the program measurably and show a few examples of how it improved the moves cosmetically, perhaps even comparing both version with specific positions. I wouldn't be opposed to fixing my program if I couldn't measure a weakening and got better looking moves in the ending. - Don > s. > > > On Thu, Mar 6, 2008 at 8:30 AM, Don Dailey <[EMAIL PROTECTED]> wrote: > >> Christoph Birk wrote: >> > On Mar 5, 2008, at 11:58 AM, Don Dailey wrote: >> >>> Don Dailey wrote: >> not assuming that MC plays the best move. The problem isn't the >> >> assumptions I am making, but the assumptions others are making, that >> >> it's NOT playing the best move.You want to apply a fix to all >> >> positions without really knowing which positions are a problem. >> > >> > One last time: Nobody suggested a one fix for all positions/problems. >> > The "floating komi" was suggested to guide the UCT search along >> > certain lines of play during specific (close!) endgame positions. >> When I said all positions I meant all games.You expect to apply this >> to all winning and losing positions in every game, not just specific ones. >> >> - Don >> >> >> >> > >> > Christoph >> > >> > ___ >> > 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: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
yes, and the fact that turning a dumpling into a dead group can take more than a few moves, since you may have to fill up the eyespace several times, meaning going fairly deeply down branches with several self-ataris along the way. s. On Thu, Mar 6, 2008 at 9:23 AM, Don Dailey <[EMAIL PROTECTED]> wrote: > > > > > > > > You won't find that in computer vs computer games, because "tricking" the > > strong programs requires some go skill and it only works if you wait long > > enough before you "solve" the position. But if you search KGS (LeelaBot, > > CrazyStone, CzechBot) for even games where the bot lost against a kyu > > players you will find many. All go more or less like that: > > > > A 4-6 kyu human is behind by 10-15 points in the midgame (at that > > stage the > > probability of winning is correlated with territory, so the MC bot is > > building fine.) He creates a 12-16 point worth nakade trick in a corner > > and does not solve it.The bot is happy, it thinks a bulk five is alive or > > something like that. Perhaps the human sacrificed another 15 points > > somewhere to create the trick so he should be dead lost. But, he only > > has to play on, reduce, etc. As the endgame approaches, the MC bot > > allows the reduction only until the territorial balance would change the > > winner. The player is happy, he turned a 25 points loss into a 1.5 point > > loss (assumed by the program) and has a 12 point surprise. > > At the end, when the whole board is decided, the player kills > > the bot's group and the bot turns a sure win into a sure loss and > > resigns. > > > > Because the trick can only be played by similar strength players (much > > weaker players can't build something like that, much stronger don't > > need it) > > it affects the rating of the bots. I guess CrazyStone could be near > > KGS 1dan > > with that solved. It is 2k now. But, of course, the solution may not > > come at > > the price of making the program weaker. That is the difficult part. > > I want to make sure I understand the nakade problem, please correct me > if I am wrong: > > My understanding of this is that many program do not allow self-atari > moves in the play-outs because in general the overwhelming majority are > stupid moves. Is that what is causing the nakade problem? And if > you start including self-atari you weaken the program in general? > > And can I assume the tree portion is also inhibited from seeing this due > to a combination of factors such as heuristics to delay exploring "ugly" > moves as well as the weakness of the play-outs in this regard (which > would cause the tree to not be inclined to get close enough to the issue > to understand it properly?) > > - Don > > > > > > > > > > > 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/ > ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Floating komi
why doesn't someone simply try this and post the results, if they think that it would help? s. On Thu, Mar 6, 2008 at 8:30 AM, Don Dailey <[EMAIL PROTECTED]> wrote: > > > Christoph Birk wrote: > > On Mar 5, 2008, at 11:58 AM, Don Dailey wrote: > >>> Don Dailey wrote: > not assuming that MC plays the best move. The problem isn't the > >> assumptions I am making, but the assumptions others are making, that > >> it's NOT playing the best move.You want to apply a fix to all > >> positions without really knowing which positions are a problem. > > > > One last time: Nobody suggested a one fix for all positions/problems. > > The "floating komi" was suggested to guide the UCT search along > > certain lines of play during specific (close!) endgame positions. > When I said all positions I meant all games.You expect to apply this > to all winning and losing positions in every game, not just specific ones. > > - Don > > > > > > > Christoph > > > > ___ > > 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: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
--- Jacques Basaldúa <[EMAIL PROTECTED]> wrote: > > Petr Baudis wrote: > > You won't find that in computer vs computer games, > because "tricking" the > strong programs requires some go skill and it only > works if you wait long > enough before you "solve" the position. But if you > search KGS (LeelaBot, > CrazyStone, CzechBot) for even games where the bot > lost against a kyu > players you will find many. All go more or less like > that: Thanks for an excellent description of "the nakade problem." I've found that it is easy for a "5kyu KGS player" - myself - to exploit such situations. I can't escape observing that endgame moves where a bot permits me to take a yose point here, another there, all the while drawing closer to a mythical 0.5 win in it's mind, lead to the position where a miscalculation about a group puts the program inescapably behind. My hypothesis is that the computer is thinking the group has a greater than 50% chance of living, but this is true only if the computer plays a vital point and/or the opponent fails to take that point. In those cases, all playouts lead to a win for the computer. But anyone about 8-10kyu is likely to spot the vital point and take it, leading to a 100% loss for the computer. >From observation, mid kyu players will set up these situations and slay the computer with a high degree of probability - say 80%. > A 4-6 kyu human is behind by 10-15 points in the > midgame (at that stage the > probability of winning is correlated with territory, > so the MC bot is > building fine.) He creates a 12-16 point worth > nakade trick in a corner > and does not solve it.The bot is happy, it thinks a > bulk five is alive or > something like that. Perhaps the human sacrificed > another 15 points > somewhere to create the trick so he should be dead > lost. But, he only > has to play on, reduce, etc. As the endgame > approaches, the MC bot > allows the reduction only until the territorial > balance would change the > winner. The player is happy, he turned a 25 points > loss into a 1.5 point > loss (assumed by the program) and has a 12 point > surprise. > At the end, when the whole board is decided, the > player kills > the bot's group and the bot turns a sure win into a > sure loss and resigns. > > Because the trick can only be played by similar > strength players (much > weaker players can't build something like that, much > stronger don't need > it) > it affects the rating of the bots. I guess > CrazyStone could be near KGS > 1dan > with that solved. It is 2k now. But, of course, the > solution may not > come at > the price of making the program weaker. That is the > difficult part. Terry McIntyre <[EMAIL PROTECTED]> Wherever is found what is called a paternal government, there is found state education. It has been discovered that the best way to insure implicit obedience is to commence tyranny in the nursery. Benjamin Disraeli, Speech in the House of Commons [June 15, 1874] Never miss a thing. Make Yahoo your home page. http://www.yahoo.com/r/hs ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
steve uurtamo wrote: > yes, and the fact that turning a dumpling into a dead > group can take more than a few moves, since you may > have to fill up the eyespace several times, meaning > going fairly deeply down branches with several self-ataris > along the way. > Ok, it's pretty much as I thought. There are relatively simple solutions but they will slow down the play-outs significantly unless someone finds a creative fast solution. I think the general outline is that you pre-test groups first to see if a self-atari move is "interesting."It's worthy of additional consideration if the stones it is touching have limited liberties and the group you self-atari is relatively small.Then you could go on to other tests which will consume even more time of course. This is probably one of those changes necessary to improve the scaling curve. The slowdown will hurt at low levels and at some point it will break even, then be stronger. - Don > s. > > On Thu, Mar 6, 2008 at 9:23 AM, Don Dailey <[EMAIL PROTECTED]> wrote: > >> > >> > >> > You won't find that in computer vs computer games, because "tricking" the >> > strong programs requires some go skill and it only works if you wait long >> > enough before you "solve" the position. But if you search KGS (LeelaBot, >> > CrazyStone, CzechBot) for even games where the bot lost against a kyu >> > players you will find many. All go more or less like that: >> > >> > A 4-6 kyu human is behind by 10-15 points in the midgame (at that >> > stage the >> > probability of winning is correlated with territory, so the MC bot is >> > building fine.) He creates a 12-16 point worth nakade trick in a corner >> > and does not solve it.The bot is happy, it thinks a bulk five is alive or >> > something like that. Perhaps the human sacrificed another 15 points >> > somewhere to create the trick so he should be dead lost. But, he only >> > has to play on, reduce, etc. As the endgame approaches, the MC bot >> > allows the reduction only until the territorial balance would change the >> > winner. The player is happy, he turned a 25 points loss into a 1.5 point >> > loss (assumed by the program) and has a 12 point surprise. >> > At the end, when the whole board is decided, the player kills >> > the bot's group and the bot turns a sure win into a sure loss and >> > resigns. >> > >> > Because the trick can only be played by similar strength players (much >> > weaker players can't build something like that, much stronger don't >> > need it) >> > it affects the rating of the bots. I guess CrazyStone could be near >> > KGS 1dan >> > with that solved. It is 2k now. But, of course, the solution may not >> > come at >> > the price of making the program weaker. That is the difficult part. >> >> I want to make sure I understand the nakade problem, please correct me >> if I am wrong: >> >> My understanding of this is that many program do not allow self-atari >> moves in the play-outs because in general the overwhelming majority are >> stupid moves. Is that what is causing the nakade problem? And if >> you start including self-atari you weaken the program in general? >> >> And can I assume the tree portion is also inhibited from seeing this due >> to a combination of factors such as heuristics to delay exploring "ugly" >> moves as well as the weakness of the play-outs in this regard (which >> would cause the tree to not be inclined to get close enough to the issue >> to understand it properly?) >> >> - Don >> >> >> >> >> >> > >> > >> > 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/ >> >> > ___ > 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: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
> I think the general outline is that you pre-test groups first to see if > a self-atari move is "interesting."It's worthy of additional > consideration if the stones it is touching have limited liberties and > the group you self-atari is relatively small.Then you could go on to > other tests which will consume even more time of course. Here is an alternative, still in the line of gathering information from play-outs. Do not turn down self-ataris TOO agressively. When a self-atari occurs in play-out: - notice which (and when). - see if the self-ataried position is yours at the end (indeed it should for a nakade). If not: cancel the play-out (or start again from the moment you self-ataried) Add a -1 to self-atarying at this place. After say 10 (-1): strictly forbid this self-atari in future playouts. If yes: Remove all constraints on this self-atari (maybe give a bonus, or study within the tree). Notice that this would also deal with seki: self-atari for both sides getting completely forbidden at one point. Jonas ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
> Thanks for an excellent description of "the nakade > problem." I've found that it is easy for a "5kyu KGS > player" - myself - to exploit such situations. I can't > escape observing that endgame moves where a bot > permits me to take a yose point here, another there, > all the while drawing closer to a mythical 0.5 win in > it's mind, lead to the position where a miscalculation > about a group puts the program inescapably behind. > How do you know this is really happening? Do you have access to what the program is thinking? Could it be possible that the program knows it is lost the whole time?That's what it seems like to me from your description.Unless you calculated the exact score considering the nakade you can't really know that this is what happened. If you have access to the program logs, you would know, but I would be willing to bet that if you think that is what is going on, the program probably is lost and knows it - because you would get the same behavior. At the very least you wouldn't be able to tell either way. To get the evil nakade behavior, you have to have several events conspire to make this happen. You have to have a nakade pattern on the board somewhere, the score has to be close and in your favor considering the nakade, and the program has to believe that it is more advantageous to give away stones that not. Despite what many people believe, MC programs don't normally believe it's better to win small and they are not hell-bent on giving away stones in order to try to make the score come out to be exactly 0.5 win. Although it's easy to see that nakade is a problem, I agree with someone who said it takes a lot of skill to produce this.In fact, I believe that it cannot be done reliably by any player unless he is already much stronger than the program, in which case he doesn't "need" to do it in order to beat the program.(Indeed, it may be a counterproductive strategy if it distracts you from playing good moves.) Most MC programs won't just let you pick off points because that is normally a strategy that decreases your winning chances. They will only do that if every move leads to the same win or loss in every single play-out, or if the small win turns out to be easier to manage. I have to believe that this is not in general a "technique" to be used to consistently beat a MC program, it is more a tool of opportunity - you can probably set it up if everything is just right to begin with and/or the program stumbles into in more or less on it's own and loses because of it. I would love to see a 5kyu player get on KGS and beat mogo in more than 1 out of 10 games using this specific strategy.My guess is that if your energy is spent setting up this trick, you will play weaker in general. - Don > My hypothesis is that the computer is thinking the > group has a greater than 50% chance of living, but > this is true only if the computer plays a vital point > and/or the opponent fails to take that point. In > those cases, all playouts lead to a win for the > computer. But anyone about 8-10kyu is likely to spot > the vital point and take it, leading to a 100% loss > for the computer. > > >From observation, mid kyu players will set up these > situations and slay the computer with a high degree of > probability - say 80%. > >> A 4-6 kyu human is behind by 10-15 points in the >> midgame (at that stage the >> probability of winning is correlated with territory, >> so the MC bot is >> building fine.) He creates a 12-16 point worth >> nakade trick in a corner >> and does not solve it.The bot is happy, it thinks a >> bulk five is alive or >> something like that. Perhaps the human sacrificed >> another 15 points >> somewhere to create the trick so he should be dead >> lost. But, he only >> has to play on, reduce, etc. As the endgame >> approaches, the MC bot >> allows the reduction only until the territorial >> balance would change the >> winner. The player is happy, he turned a 25 points >> loss into a 1.5 point >> loss (assumed by the program) and has a 12 point >> surprise. >> At the end, when the whole board is decided, the >> player kills >> the bot's group and the bot turns a sure win into a >> sure loss and resigns. >> >> Because the trick can only be played by similar >> strength players (much >> weaker players can't build something like that, much >> stronger don't need >> it) >> it affects the rating of the bots. I guess >> CrazyStone could be near KGS >> 1dan >> with that solved. It is 2k now. But, of course, the >> solution may not >> come at >> the price of making the program weaker. That is the >> difficult part. >> > > > Terry McIntyre <[EMAIL PROTECTED]> > > “Wherever is found what is called a paternal government, there is found state > education. It has been discovered that the best way to insure implicit > obedience is to commence tyranny in the nursery.” > > Benjamin Disraeli, Speech in the House of Commons [J
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
Your idea is more in the spirit of MC, I like it. Another idea is borrowed from my first reasonable MC player. I looked at the "futures" of interesting move points and discouraged self-atari moves unless the future belonged to the player executing the move. (A "future" is the expected percentage of time a given player ended up with a given point at the end of the random games.) So some sort of pre-processed quick all-moves-as-first random play-out can give you a sense of which self-atari points are interesting. But it is not dynamic unfortunately and thus not scalable unless done periodically during the tree search. - Don Jonas Kahn wrote: >> I think the general outline is that you pre-test groups first to see if >> a self-atari move is "interesting."It's worthy of additional >> consideration if the stones it is touching have limited liberties and >> the group you self-atari is relatively small.Then you could go on to >> other tests which will consume even more time of course. >> > > > Here is an alternative, still in the line of gathering information from > play-outs. > > Do not turn down self-ataris TOO agressively. > When a self-atari occurs in play-out: > - notice which (and when). > - see if the self-ataried position is yours at the end (indeed it should > for a nakade). > If not: cancel the play-out (or start again from the moment you > self-ataried) Add a -1 to self-atarying at this place. > After say 10 (-1): strictly forbid this self-atari in future > playouts. > If yes: Remove all constraints on this self-atari (maybe give a bonus, > or study within the tree). > > Notice that this would also deal with seki: self-atari for both sides > getting completely forbidden at one point. > > Jonas > ___ > 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: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
Because the trick can only be played by similar strength players (much weaker players can't build something like that, much stronger don't need it) it affects the rating of the bots. I guess CrazyStone could be near KGS 1dan with that solved. It is 2k now. But, of course, the solution may not come at the price of making the program weaker. That is the difficult part. I'm a 6k EGF player and I manage to win about 80% of the time against mogo and Leela on a [EMAIL PROTECTED] with 120min. (I only played 30 games each, so it may be a fluke, but...). If you ask me - on 19x19 dan status is far away. I think that Nakade will have to be solved before anything else. mfg Florian Erhardt ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
On Thu, Mar 06, 2008 at 12:55:53PM +, Jacques Basaldúa wrote: > A 4-6 kyu human is behind by 10-15 points in the midgame (at that stage the > probability of winning is correlated with territory, so the MC bot is > building fine.) He creates a 12-16 point worth nakade trick in a corner > and does not solve it.The bot is happy, it thinks a bulk five is alive or > something like that. Perhaps the human sacrificed another 15 points > somewhere to create the trick so he should be dead lost. But, he only > has to play on, reduce, etc. As the endgame approaches, the MC bot > allows the reduction only until the territorial balance would change the > winner. The player is happy, he turned a 25 points loss into a 1.5 point > loss (assumed by the program) and has a 12 point surprise. > At the end, when the whole board is decided, the player kills > the bot's group and the bot turns a sure win into a sure loss and resigns. I honestly don't think at all that these "tricks" are created by the opponents meaningfully, in 90% of the cases I think they arise from perfectly natural corner situations; the nakade weakness is not that well known, I believe. > Because the trick can only be played by similar strength players (much > weaker players can't build something like that, much stronger don't need > it) > it affects the rating of the bots. I guess CrazyStone could be near KGS > 1dan > with that solved. It is 2k now. But, of course, the solution may not come > at > the price of making the program weaker. That is the difficult part. (Note that I believe CrazyStone played way too few games to have a precise rank. I think it could still be 3k and it could still be 1d as it is now, if it played more games. CzechBot has played thousands of games now I think, and its rank is _still_ evolving, though I don't think it will reach 2k.) -- Petr "Pasky" Baudis Whatever you can do, or dream you can, begin it. Boldness has genius, power, and magic in it.-- J. W. von Goethe ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
> You have to have a nakade pattern on the > board somewhere, the score has to be close and in your favor > considering the nakade, and the program has to believe that it is more > advantageous to give away stones that not. eh, or it can't see the capture until it's only a few moves away, because its horizon with respect to self-atari is so shallow. deepen the horizon and it'll consider those moves early enough not to screw up its overall win percentage evaluation. > Although it's easy to see that nakade is a problem, I agree with > someone who said it takes a lot of skill to produce this.In fact, I > believe that it cannot be done reliably by any player unless he is > already much stronger than the program, in which case he doesn't "need" > to do it in order to beat the program. I wouldn't go this far -- humans learn from their mistakes, but can stay at the same skill level regardless of how much they learn, either because they forget things that they earlier learned, or because they have very shallow reading, say. for a computer, though, it's quite possible that every single player ranked one or two stones lower than (arbitrary mc program with this weakness -- AMCW) could exploit this weakness in a systematic way, more than 50% of the time. this would eventually reduce AMCW's ranking, of course, but wouldn't raise any of those player's rankings, because their ability to beat one specific player consistently isn't enough to modify their ranking. > (Indeed, it may be a > counterproductive strategy if it distracts you from playing good moves.) these aren't bad moves in any way. they're normal, healthy, strong go-player moves that are recognized instantly by anyone who has a read a copy of "life and death" or similar. > Most MC programs won't just let you pick off points because that is > normally a strategy that decreases your winning chances. They will > only do that if every move leads to the same win or loss in every single > play-out, or if the small win turns out to be easier to manage. or if they aren't reading out the playouts deeply enough that would allow them to correctly consider the impact of those moves early enough to avoid them! > I would love to see a 5kyu player get on KGS and beat mogo in more than > 1 out of 10 games using this specific strategy.My guess is that if > your energy is spent setting up this trick, you will play weaker in general. i dunno. imagine one of the "mate in 20" types of sequences that you're supposed to learn when you first learn chess. imagine that you never learn how to deal with them. s. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Floating komi
On Thu, 6 Mar 2008, Don Dailey wrote: One last time: Nobody suggested a one fix for all positions/problems. The "floating komi" was suggested to guide the UCT search along certain lines of play during specific (close!) endgame positions. When I said all positions I meant all games.You expect to apply this to all winning and losing positions in every game, not just specific ones. No. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
On Thu, 6 Mar 2008, Don Dailey wrote: And can I assume the tree portion is also inhibited from seeing this due to a combination of factors such as heuristics to delay exploring "ugly" moves as well as the weakness of the play-outs in this regard (which would cause the tree to not be inclined to get close enough to the issue to understand it properly?) It might also be reading-depth. Some nakade forms need quite deep reading you want to "discover" them on-the-fly. 10 kyu humans know that the bulky-five is dead; no reading required. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
Don Dailey wrote: > Although it's easy to see that nakade is a problem, I agree with > someone who said it takes a lot of skill to produce this. In fact, I > believe that it cannot be done reliably by any player unless he is > already much stronger than the program, in which case he doesn't > "need" to do it in order to beat the program. It turns out not to be difficult at all. When it has a killable group, Mogo actively cooperates to bring about these situations, because it thinks it has found a way to live. I have killed more groups in bent-4 in a few evenings of playing mogo at 13x13 than I have otherwise in the last ten years. > Most MC programs won't just let you pick off points because that is > normally a strategy that decreases your winning chances. They will > only do that if every move leads to the same win or loss in every > single play-out, or if the small win turns out to be easier to manage. If you watch mogo playing on KGS, you will see that when it ahead it does consistently let its opponent 'pick off points' until the game becomes very close. -M- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
On Thu, 6 Mar 2008, Don Dailey wrote: advantageous to give away stones that not. Despite what many people believe, MC programs don't normally believe it's better to win small and they are not hell-bent on giving away stones in order to try to make the score come out to be exactly 0.5 win. You are correct that it is not explicitly programmed into the MC programs to win by 0.5 pts, but since most of them don't care about the margin they in practice often do. As you might remember I (3 kyu) played many games on CGOS and many games that I lost, I actually lost by 0.5 pts. It mostly worked like this: I am behind by several points in the early endgame, then the programs allow me to gain a point here, or there. But in the end they still win by 0.5 pts. These programs are NOT "hell-bent on losing", they just dont care if the UCT-tree shows that they win anyway. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Floating komi
Don Dailey wrote: > I would be satisfied if someone implemented it, reported a 500 game > self-test sample and concluded that it didn't hurt the program > measurably and show a few examples of how it improved the moves > cosmetically, perhaps even comparing both version with specific > positions. Remember that there's good reason to believe that this technique will be less helpful in self-play. -M- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
On 3/6/08, Christoph Birk <[EMAIL PROTECTED]> wrote: > On Thu, 6 Mar 2008, Don Dailey wrote: > > > advantageous to give away stones that not. Despite what many people > > believe, MC programs don't normally believe it's better to win small > > and they are not hell-bent on giving away stones in order to try to make > > the score come out to be exactly 0.5 win. > > > You are correct that it is not explicitly programmed into the MC > programs to win by 0.5 pts, but since most of them don't care about > the margin they in practice often do. You are right, but I think that you may also be misconstruing the nakade problem as a lack of concern about margin, when it is really a fundamental failure to understand (i.e., failure to explore sufficiently) the nakade shapes. The tendancy for your final scores on lost games to be 0.5 really reflects a motivation on your own part to lose by the smallest margin that you can. After a point when the game is resolved, this doesn't conflict with the program's goal, so it lets you do just that. As an interesting thought, I think that it might actually be informative for people to try out programs that _do_ try to win by exactly 0.5! Not as a fundamental goal, but rather as a slight preference for the endgame. If authors can do this in a manner that does not degrade the playing ability much, then human players might be able to see even more clearly what life & death errors a program makes, and at what point it is able to discover the correct solution. Clearly the programs would be weaker like this, but I think it could expose some systematic problems very quickly, so that they could be focused on. Weston ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: endgame (Was [computer-go] Re: Should 9x9 komi be 8.0 ?])
On Thu, 6 Mar 2008, Weston Markham wrote: You are right, but I think that you may also be misconstruing the nakade problem as a lack of concern about margin, when it is really a fundamental failure to understand (i.e., failure to explore Sorry, you miss-understood. The nakade problem is totally unrelated to the margin problem. They just sometimes happen at the same time and then allow someone to take advantage of them. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: komi argument = silly
To a first order approximation, would changing the komi change the rankings? Presumably, programs are playing the same number of games as black and white, so any "unfair" advantage or disadvantage black has would balance out. Komi only matters when there is only one game between a pair of opponents. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: komi argument = silly
What komi would do is push the ratings closer together, but it wouldn't change the ranks of the players. - Don Dave Dyer wrote: > To a first order approximation, would changing the komi change the > rankings? Presumably, programs are playing the same number of games > as black and white, so any "unfair" advantage or disadvantage black > has would balance out. > > Komi only matters when there is only one game between a pair of opponents. > > ___ > 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: komi argument = silly
On Thu, Mar 06, 2008 at 04:33:16PM -0800, Dave Dyer wrote: > > To a first order approximation, would changing the komi change the > rankings? Presumably, programs are playing the same number of games > as black and white, so any "unfair" advantage or disadvantage black > has would balance out. > > Komi only matters when there is only one game between a pair of opponents. This has nothing to do with black/white distinction. The idea is to dynamically adjust the komi to make UCT to aim at higher and potentially less sure win or lower and potentially more sure loss. Of course, depending on whether it takes black or white you would adjust the komi in the correct direction. -- Petr "Pasky" Baudis Whatever you can do, or dream you can, begin it. Boldness has genius, power, and magic in it.-- J. W. von Goethe ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] there may be life left in traditional programs yet
Often, when I study sprawling groups in the middle game, I have found that gnugo --decide-dragon-status will fail with an uncertain result, but if I increase the owl-node-limits and semeai-node-limits to 10k, gnugo finds a resolution to the problem in a matter of seconds. I sall run gnugo's solutions past stronger players, but at the moment, they look reasonable to me; certainly they are playable against low kyu opponents. I suspect that gnugo's limits were tuned for slower processors and smaller memory sets. Now that machines come "off the shelf" with 3 gigabytes of RAM, perhaps it's time to revisit those parameters. A 10 megabyte cache ( the default ) seems too parsimonious. If traditional programs were to fully use the RAM now available, building trees with tens or hundreds of thousands of nodes, it looks to me like their middle game on a 19x19 board might impress dan-level players. Terry McIntyre <[EMAIL PROTECTED]> Wherever is found what is called a paternal government, there is found state education. It has been discovered that the best way to insure implicit obedience is to commence tyranny in the nursery. Benjamin Disraeli, Speech in the House of Commons [June 15, 1874] Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
