Re: [computer-go] Re: Hahn system tournament and MC bots
Hi all, If I may get out of lurking mode and try to understand the problem here... IMHO there is another issue here that creates a difference and makes the strategies for "normal go" and "hahn go" incomparable. I has been touched upon by previous posters, but not spelled out. Normal go strategy applies as-is to both a single game and to a whole tournament. One plays the in the same way no matter what happens outside the current game. Hahn go strategy is only relevant for a tournament (otherwise one can simply play normal go, it doesn't matter by how many points one wins). And thus it includes a meta-strategy involving the results in the other games and knowledge of one's opponents. Using a hahn strategy in a single game or in a tournament that is scored as normal go will probably yield worse performance, like it was pointed out here before, because it requires more processing whose results are then ignored. Or am I missing something? best regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
Hi, > Hahn go strategy is only relevant for a tournament (otherwise one can > simply play normal go, it doesn't matter by how many points one wins). > And thus it includes a meta-strategy involving the results in the > other games and knowledge of one's opponents. > One can also play a single game for instance with money bets based on the Hahn points, which makes Hahn go strategy relevant also for a single game. In the tournament setting, in your interpretation, the goal is not to maximize the (expected) number of Hahn points in each game, but to maximize the probability of having more Hahn points at the end of the tournament than your opponent(s). It would also be useful to see what is happening on the other boards during a tournament round, since it might affect your point goal. It might even be useful to spend time waiting in order to gather information from the other boards. ;-) Tapani -- Tapani Raiko, , +358 50 5225750 http://www.iki.fi/raiko/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 11:18, Tapani Raiko wrote: > Hi, >> Hahn go strategy is only relevant for a tournament (otherwise one can >> simply play normal go, it doesn't matter by how many points one wins). >> And thus it includes a meta-strategy involving the results in the >> other games and knowledge of one's opponents. >> > One can also play a single game for instance with money bets based on > the Hahn points, which makes Hahn go strategy relevant also for a single > game. Right. Didn't think about that. The idea however is that it's another game altogether than normal go and we really can't compare playing strength between them. > In the tournament setting, in your interpretation, the goal is not to > maximize the (expected) number of Hahn points in each game, but to > maximize the probability of having more Hahn points at the end of the > tournament than your opponent(s). It would also be useful to see what is > happening on the other boards during a tournament round, since it might > affect your point goal. It might even be useful to spend time waiting in > order to gather information from the other boards. ;-) Exactly. The goal is to win the tournament and this involves more than just playing games. The meta-strategy is probably independent of which games are played, but it affects the strategy chosen in each instance of a game. Of course, one can just ignore all that and just win all games with 40 points or more. :-) Then one can use a normal go strategy with the appropriate komi. regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 11:18, Tapani Raiko wrote: > One can also play a single game for instance with money bets based on > the Hahn points, which makes Hahn go strategy relevant also for a single > game. Just a thought: if the bet is "I can beat you with X points on the board or more", then it's exactly like trying to win a normal game with X points komi, right? Are there any other kind of bets? regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
Vlad Dumitrescu wrote: > On Tue, Nov 24, 2009 at 11:18, Tapani Raiko wrote: > >> One can also play a single game for instance with money bets based on >> the Hahn points, which makes Hahn go strategy relevant also for a single >> game. >> > > Just a thought: if the bet is "I can beat you with X points on the > board or more", then it's exactly like trying to win a normal game > with X points komi, right? > > Are there any other kind of bets? > Yes, having to pay the amount of Hahn points in money. The Hahn system originates from the Korean betting system, mentioned also in the novel First Kyu by Sung-Hwa Hong. Both players deposit the amount for the maximum loss under the go board and the money is split after the game according to the score. See also: http://www.suomigo.net/wiki/HahnSystem Tapani -- Tapani Raiko, , +358 50 5225750 http://www.iki.fi/raiko/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 14:20, Tapani Raiko wrote: > Vlad Dumitrescu wrote: >> Just a thought: if the bet is "I can beat you with X points on the >> board or more", then it's exactly like trying to win a normal game >> with X points komi, right? >> >> Are there any other kind of bets? >> > Yes, having to pay the amount of Hahn points in money. The Hahn system > originates from the Korean betting system, mentioned also in the novel > First Kyu by Sung-Hwa Hong. Both players deposit the amount for the > maximum loss under the go board and the money is split after the game > according to the score. Thanks, I had read that. So the only difference in play is when losing, one has to keep trying to lose as little as possible, resigning isn't an option. When ahead, there's no reason to try to win big, unless the goal is to reach a certain amount of points in a certain number of games. (Programs aren't greedy in the same way we people are :-) regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 03:06:51PM +0100, Vlad Dumitrescu wrote: > > So the only difference in play is when losing, one has to keep trying > to lose as little as possible, resigning isn't an option. When ahead, > there's no reason to try to win big, unless the goal is to reach a > certain amount of points in a certain number of games. (Programs > aren't greedy in the same way we people are :-) Let's assume that the program will play for a gambler, and will play many (an indefinite number) of independent games. Then I think "no reason to try to win big" is wrong. The rational approach to gambling is to maximize your expected value for each game. So now the problem becomes harder -- you have to realistically guess the risk vs reward over a spectrum of points. I think this game is clearly more difficult than a binary win/loss game. -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 15:45, Jeff Nowakowski wrote: > On Tue, Nov 24, 2009 at 03:06:51PM +0100, Vlad Dumitrescu wrote: >> So the only difference in play is when losing, one has to keep trying >> to lose as little as possible, resigning isn't an option. When ahead, >> there's no reason to try to win big, unless the goal is to reach a >> certain amount of points in a certain number of games. (Programs >> aren't greedy in the same way we people are :-) > > Let's assume that the program will play for a gambler, and will play > many (an indefinite number) of independent games. Then I think "no > reason to try to win big" is wrong. The rational approach to gambling > is to maximize your expected value for each game. Yes, but that doesn't necessarily mean that the strategy should be to push each game to the limit. Trying to win with a large margin is less safe than with a small one, so it depends on the gambler's mindset. > So now the problem > becomes harder -- you have to realistically guess the risk vs reward > over a spectrum of points. And also possibly add knowledge about the opponent(s). > I think this game is clearly more difficult than a binary win/loss > game. This is exactly my opinion, too. regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 03:57:37PM +0100, Vlad Dumitrescu wrote: > > Yes, but that doesn't necessarily mean that the strategy should be to > push each game to the limit. Trying to win with a large margin is less > safe than with a small one, so it depends on the gambler's mindset. That's why I said expected value, and specified an indefinite number of independent games. In this case, the only rational play is to judge the winning percentage for each move and weight it against the value. Mindset has nothing to do with it in this situation. This is standard gambling behavior for this kind of game. > And also possibly add knowledge about the opponent(s). This problem is applicable to the standard win/loss game too -- programs are sometimes tuned against other programs. For the sake of discussion, I'm assuming that you know nothing about the other player and just assume best play from him. -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
In message <95be1d3b0911240657g24467ecey84cdb05918ca7...@mail.gmail.com>, Vlad Dumitrescu writes On Tue, Nov 24, 2009 at 15:45, Jeff Nowakowski wrote: On Tue, Nov 24, 2009 at 03:06:51PM +0100, Vlad Dumitrescu wrote: So the only difference in play is when losing, one has to keep trying to lose as little as possible, resigning isn't an option. When ahead, there's no reason to try to win big, unless the goal is to reach a certain amount of points in a certain number of games. (Programs aren't greedy in the same way we people are :-) Let's assume that the program will play for a gambler, and will play many (an indefinite number) of independent games. Then I think "no reason to try to win big" is wrong. The rational approach to gambling is to maximize your expected value for each game. Yes, but that doesn't necessarily mean that the strategy should be to push each game to the limit. Trying to win with a large margin is less safe than with a small one, so it depends on the gambler's mindset. It's not a matter of "pushing .. to the limit", it's a matter of maximising expected winnings. And it shouldn't depend on mindset, it should depend on calculation. Suppose my attempts to read the game tell me "If I seal off my territory at A, I will win by 5 points. If instead I invade at B, then 70% of the time I will win by 25 points, 30% of the time I will lose by 5 points". If I am playing Go, I will prefer A. If I am playing bang neki, I will prefer B. Nick So now the problem becomes harder -- you have to realistically guess the risk vs reward over a spectrum of points. And also possibly add knowledge about the opponent(s). I think this game is clearly more difficult than a binary win/loss game. This is exactly my opinion, too. regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- Nick Weddn...@maproom.co.uk ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 16:11, Nick Wedd wrote: > Suppose my attempts to read the game tell me "If I seal off my territory at > A, I will win by 5 points. If instead I invade at B, then 70% of the time I > will win by 25 points, 30% of the time I will lose by 5 points". > > If I am playing Go, I will prefer A. If I am playing bang neki, I will > prefer B. Sure. But different gamblers have different "break-even" limits, i.e. different mindsets. Some are cautious and prefer 80% for those 25 points; some are reckless and would go for B even with 60%. Anyway, we are digressing. Thank you all for the response! regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 04:19:45PM +0100, Vlad Dumitrescu wrote: > > Sure. But different gamblers have different "break-even" limits, i.e. > different mindsets. Some are cautious and prefer 80% for those 25 > points; some are reckless and would go for B even with 60%. No professional gambler, if he had the numbers laid out for him, would ever choose unoptimal play, not when he's playing for the long term. The computer, in the same way, would have to be modeled to maximize expected value. Nothing else makes sense. In a single game with high stakes, yes mindset (risk aversion, your finances, etc) might come into it. But that is exactly why I specified the long term scenario. -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Optimizing combinations of flags
Brian Sheppard wrote: > I think that I am assuming only that the objective function is convex. The > parameters in Go programs are always inter-dependent. What do you do when you add a new parameter? Do you retain your existing 'history', considering each game to have been played with the value of the new parameter set to zero? If you have 50 parameters already, doesn't adding a new parameter create a rather large number of new parameter sets, most of which there will never be time to investigate? I have been using UCB and UCT to tune engine settings, but I don't think these methods work well to tune more than a handful of parameters at the same time. -M- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 16:11, Nick Wedd wrote: > Suppose my attempts to read the game tell me "If I seal off my territory at > A, I will win by 5 points. If instead I invade at B, then 70% of the time I > will win by 25 points, 30% of the time I will lose by 5 points". > > If I am playing Go, I will prefer A. If I am playing bang neki, I will > prefer B. I think opponent modeling should not be ignored or abstracted away. The bot can estimate the probabilities for A and B, assuming an equal strength opponent. But then there is an extra step. With confidence X, the opponent is so much weaker or stronger, in which case the probabilities would be different. If the opponent has been playing random-looking moves, the smart estimate of the probabilities will be different than if it has been playing strong moves and pulling ahead. It's interesting to consider the problem of writing an agent to make side bets. There could be a pool of spectator bot, each calculating an estimate of the final score, after every move, and placing wagers. Suppose we were to hold a tournament to test out some of these theories. For CGOS, bots come and go at their minders' whim. I don't see a good way to hold a Hahn tournament like that. For KGS, perhaps it could be run as a round-robin tournament and then the scores calculated offline with a spreadsheet program. That's not ideal either. - Dave Hillis ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: A cluster version of Zen is running on cgos 19x19
Hideki Kato wrote: > I'm now testing a cluster version of Zen (Zengg-4x4c-tst), developed > by a joint project with Yamato, on cgos 19x19. It wons, however, all > games (except first one with timeout due to a bug). Running more > strong programs are very appreciated. Hideki, thx for your activity. Do I have a Christmas wish for free already? It is: Let the cluster also run on KGS - against the humans. Ingo. -- GRATIS für alle GMX-Mitglieder: Die maxdome Movie-FLAT! Jetzt freischalten unter http://portal.gmx.net/de/go/maxdome01 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: Hahn system tournament and MC bots
Jeff Nowakowski wrote: > I think this game [go with Hahn scoring; IA] is clearly more > difficult than a binary win/loss game. That is one of the possible question, and I also vote for "yes", as normal go is simply a Hahn-Go veriant with "coarsened" evaluation. Even more interesting might be this question: Assume, you have humans and bots, all of same strength in traditional go. Which of the groups will be better in Hahn-Go ? Dave Hillis proposed: > It's interesting to consider the problem of writing an agent > to make side bets. There could be a pool of spectator bot, > each calculating an estimate of the final score, after every > move, and placing wagers. I like this idea, even in the generalized version, where humans and bots are allowed to bet on the outcome(s). Ingo. -- Jetzt kostenlos herunterladen: Internet Explorer 8 und Mozilla Firefox 3.5 - sicherer, schneller und einfacher! http://portal.gmx.net/de/go/chbrowser ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: A cluster version of Zen is running on cgos 19x19
Ingo Althöfer: <20091124190802.303...@gmx.net>: >Hideki Kato wrote: >> I'm now testing a cluster version of Zen (Zengg-4x4c-tst), developed >> by a joint project with Yamato, on cgos 19x19. It wons, however, all >> games (except first one with timeout due to a bug). Running more >> strong programs are very appreciated. > >Hideki, thx for your activity. > >Do I have a Christmas wish for free already? >It is: Let the cluster also run on KGS - against the humans. I'd like to do so but it's not allowed to connect the cluster to the Internet, sigh. Hideki -- g...@nue.ci.i.u-tokyo.ac.jp (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: A cluster version of Zen is running on cgos 19x19
Hideki replied: > >> Do I have a Christmas wish for free already? >> It is: Let the cluster also run on KGS - against the humans. > > I'd like to do so but it's not allowed to connect the > cluster to the Internet, sigh. Hmm. As CGOS is also Internet, it seems that Zen-author does not allow you to connect to KGS. Is Zen-Author reading here? Maybe, he can rethink about the possibility. I want Cluster-Zen for Christmas, Cluster-Zen-for Christmas, Cluster-Zen for Christmas, please, please, please, please... Little child In-Go. -- GRATIS für alle GMX-Mitglieder: Die maxdome Movie-FLAT! Jetzt freischalten unter http://portal.gmx.net/de/go/maxdome01 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: A cluster version of Zen is running on cgos 19x19
Ingo Althöfer: <20091124200643.255...@gmx.net>: >Hideki replied: >> >>> Do I have a Christmas wish for free already? >>> It is: Let the cluster also run on KGS - against the humans. >> >> I'd like to do so but it's not allowed to connect the >> cluster to the Internet, sigh. > >Hmm. As CGOS is also Internet, it seems that Zen-author >does not allow you to connect to KGS. Ah, I was confusing. I wrote about T2K HPC cluster, which is the main target of my development, not my home cluaster. My mini cluster can freely be connected to KGS, though I have no rated bot account yet. >Is Zen-Author reading here? >Maybe, he can rethink about the possibility. He is sleeping now 'cause it's 5:30 am in Japan :). >I want Cluster-Zen for Christmas, Cluster-Zen-for Christmas, >Cluster-Zen for Christmas, please, please, please, please... > >Little child In-Go. I'll throw it into KGS after tuning several parameters. Hideki -- g...@nue.ci.i.u-tokyo.ac.jp (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: A cluster version of Zen is running on cgos 19x19
Hi Hideki, >> Is Zen-Author reading here? >> Maybe, he can rethink about the possibility. > > He is sleeping now 'cause it's 5:30 am in Japan :). Ok, let him his good sleep. >> I want Cluster-Zen for Christmas, Cluster-Zen-for Christmas, >> Cluster-Zen for Christmas, please, please, please, please... >> >> Little child In-Go. > > I'll throw it into KGS after tuning several parameters. You are a 100-%-darling. Thanks a lot in advance. Ingo. -- Jetzt kostenlos herunterladen: Internet Explorer 8 und Mozilla Firefox 3.5 - sicherer, schneller und einfacher! http://portal.gmx.net/de/go/atbrowser ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
In message <20091124193826.303...@gmx.net>, Ingo Althöfer <3-hirn-ver...@gmx.de> writes Jeff Nowakowski wrote: I think this game [go with Hahn scoring; IA] is clearly more difficult than a binary win/loss game. That is one of the possible question, and I also vote for "yes", as normal go is simply a Hahn-Go veriant with "coarsened" evaluation. But the "additive" property of Hahn scoring makes life easy for players. If the board has become separated into regions that do not interact, players can just work out what they think is the biggest local move on each part of the board, and then make the biggest of these moves. This calculation is correct for Hahn scoring, but not for normal scoring, or indeed any other way of scoring. (I am not talking about tedomari effects here, which are rare and small; I am talking about the handling of uncertainty.) In fact, I believe that of all the ways of converting from the board score to the object of the game, Hahn scoring is the uniquely easiest and least interesting. Even more interesting might be this question: Assume, you have humans and bots, all of same strength in traditional go. Which of the groups will be better in Hahn-Go ? The humans. All the currently strongest bots use some form of MC UCT, which works surprisingly well for traditional Go and is not intended for Hahn Go. Most humans below around 5-kyu play traditional Go as if they were playing Hahn Go (because they can't count, or because they have never learned to apply the results of their counting). So I expect humans at the same level as leading bots, i.e. around 1-dan, can still manage to adjust their play for Hahn Go if asked to. Nick Dave Hillis proposed: It's interesting to consider the problem of writing an agent to make side bets. There could be a pool of spectator bot, each calculating an estimate of the final score, after every move, and placing wagers. I like this idea, even in the generalized version, where humans and bots are allowed to bet on the outcome(s). Ingo. -- Nick Weddn...@maproom.co.uk ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: A cluster version of Zen is running on cgos 19x19
In message <4b0c4522.370%hideki_ka...@ybb.ne.jp>, Hideki Kato writes Ingo Althöfer: <20091124200643.255...@gmx.net>: Hideki replied: Do I have a Christmas wish for free already? It is: Let the cluster also run on KGS - against the humans. I'd like to do so but it's not allowed to connect the cluster to the Internet, sigh. Hmm. As CGOS is also Internet, it seems that Zen-author does not allow you to connect to KGS. Ah, I was confusing. I wrote about T2K HPC cluster, which is the main target of my development, not my home cluaster. My mini cluster can freely be connected to KGS, though I have no rated bot account yet. Is Zen-Author reading here? Maybe, he can rethink about the possibility. He is sleeping now 'cause it's 5:30 am in Japan :). I want Cluster-Zen for Christmas, Cluster-Zen-for Christmas, Cluster-Zen for Christmas, please, please, please, please... Little child In-Go. I'll throw it into KGS after tuning several parameters. The December KGS bot tournament will be 9x9. I guess that if a cluster-Zen competes in that (I am hoping it will), it will be unbeatable. The existing pattern of KGS bot tournaments (see http://www.weddslist.com/kgs/future.html) means that the January one will also be 9x9, then February and March will both be 19x19. A cluster Zen in a 19x19 event will be even more interesting to watch. Nick Hideki -- g...@nue.ci.i.u-tokyo.ac.jp (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- Nick Weddn...@maproom.co.uk ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 22:15, Nick Wedd wrote: > But the "additive" property of Hahn scoring makes life easy for players. If > the board has become separated into regions that do not interact, players > can just work out what they think is the biggest local move on each part of > the board, and then make the biggest of these moves. This calculation is > correct for Hahn scoring, but not for normal scoring, or indeed any other > way of scoring. (I am not talking about tedomari effects here, which are > rare and small; I am talking about the handling of uncertainty.) > > In fact, I believe that of all the ways of converting from the board score > to the object of the game, Hahn scoring is the uniquely easiest and least > interesting. I'm sorry to bother you, but I don't get it. There must be some subtle detail that escapes me... Please try to explain why the "hahn calculation" isn't working in a normal game so as to ensure a win. I'm talking about strong human players. In my view, we have hahn: object of the game = max board score normal: object of the game = board score > komi Both seem just as easy and interesting. best regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots. and KGS tournament ?
Le 24/11/2009 à 00:24, dhillism...@netscape.net a écrit : > > For my fast/dumb neural net engine, Antbot9x9, I coevolved the weights using > a similar tournament system. Each individual played a number of games against > all the others, round robin, and the score was the sum of points for all of > its games. > > Some observations/claims: > Non-transitive effects seem more visible. Consistently overplaying garners > extra points from weak opponents but needlessly loses extra points against > strong ones. It becomes more important to play your opponent as well as the > board: if you think that you have him outmatched, take some risky gambles, > overplay. Every game in the tournament matters, right till the end of that > game. > > I think it could be interesting to try some bot tournaments like this. It > might be fun to watch. When the strongest bot was playing the weakest, even > near the (painfully one-sided) end of the game there would be an element of > suspense. The stronger bot would (or should) be trying to swindle a few last > extra points it didn't deserve, and the fate of the tournament could hinge on > it. > > - Dave Hillis > In another thread Nick Wedd wrote: > The December KGS bot tournament will be 9x9. I guess that if a > cluster-Zen competes in that (I am hoping it will), it will be > unbeatable. > > The existing pattern of KGS bot tournaments (see > http://www.weddslist.com/kgs/future.html) means that the January one > will also be 9x9, then February and March will both be 19x19. > ... Is there a possibility for an Hahn tournament on KGS ? maybe with simplified rules: one point on board is one point in tournament ( (c) R.Jasiek ) If i understand what D.Hillis said, it can put in light some hidden aspects of the bots, and should be more spectacular than the wise-sure-win style of MC *Go* bots. And i guess it does not require lot of change in the code, "only" points instead of win/loss in the evaluation function should do the trick. I hope several strong programmers would like to participate, for fun and maybe discover several things in their code by pushing it to unusual limits. Alain. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: A cluster version of Zen is running on cgos 19x19
Hi Nick, I'll perticipate comming tournaments as much as possible but it's still under development and needs much more work and time for full performance. Since my mini cluster uses usual Gigabit Ether, which is much slower than expensive Infiniband or such high speed network devices, it performs not so better on 9x9. So please do not expect much :). Also, on 19x19 board, current 16-core cluster version performs almost the same as 8-core shared memory pc such as Mac Pro, which Yamato used for KGS. Hideki Nick Wedd: : >In message <4b0c4522.370%hideki_ka...@ybb.ne.jp>, Hideki Kato > writes >> >>Ingo Althöfer: <20091124200643.255...@gmx.net>: >>>Hideki replied: > Do I have a Christmas wish for free already? > It is: Let the cluster also run on KGS - against the humans. I'd like to do so but it's not allowed to connect the cluster to the Internet, sigh. >>> >>>Hmm. As CGOS is also Internet, it seems that Zen-author >>>does not allow you to connect to KGS. >> >>Ah, I was confusing. I wrote about T2K HPC cluster, which is the main >>target of my development, not my home cluaster. My mini cluster can >>freely be connected to KGS, though I have no rated bot account yet. >> >>>Is Zen-Author reading here? >>>Maybe, he can rethink about the possibility. >> >>He is sleeping now 'cause it's 5:30 am in Japan :). >> >>>I want Cluster-Zen for Christmas, Cluster-Zen-for Christmas, >>>Cluster-Zen for Christmas, please, please, please, please... >>> >>>Little child In-Go. >> >>I'll throw it into KGS after tuning several parameters. > >The December KGS bot tournament will be 9x9. I guess that if a >cluster-Zen competes in that (I am hoping it will), it will be >unbeatable. > >The existing pattern of KGS bot tournaments (see >http://www.weddslist.com/kgs/future.html) means that the January one >will also be 9x9, then February and March will both be 19x19. A cluster >Zen in a 19x19 event will be even more interesting to watch. > >Nick > >>Hideki >>-- >>g...@nue.ci.i.u-tokyo.ac.jp (Kato) >>___ >>computer-go mailing list >>computer-go@computer-go.org >>http://www.computer-go.org/mailman/listinfo/computer-go/ -- g...@nue.ci.i.u-tokyo.ac.jp (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots. and KGS tournament ?
Alain Baeckeroot wrote: > If i understand what D.Hillis said, it can put in light some hidden > aspects of the bots, and should be more spectacular than the > wise-sure-win style of MC *Go* bots. > And i guess it does not require lot of change in the code, "only" > points instead of win/loss in the evaluation function should do the > trick. In principle it's not difficult to change from tracking win/loss counts to tracking average score + number of games. But I think it's quite common to express 'knowledge' by using rules like "if such-and-such a pattern matches, add 10 RAVE wins to the node"; it's not so obvious how to modify these rules for a "Hahn" engine. (I suppose working out a good approach to this might suggest ideas which could be used in standard engines: for example, it might turn out that rules like the one above are weighting the knowledge too low/high in positions where the engine is already doing well/badly.) -M- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
From: Vlad Dumitrescu >I'm sorry to bother you, but I don't get it. There must be some subtle >detail that escapes me... >Please try to explain why the "hahn calculation" isn't working in a >normal game so as to ensure a win. I'm talking about strong human >players. >In my view, we have >hahn: object of the game = max board score >normal: object of the game = board score > komi >Both seem just as easy and interesting. If you are winning in the Hahn sense, your score also exceeds komi; but Hahn scoring - either by accumulating points in a tournament ranking, or converting points to dollars in bang neki fashion, gives you incentive to achieve larger scores. Under the "board score > komi" regime, if you have a group which might be invaded (at some risk of losing points), but which can be safely walled off, I might choose to wall it off if my overall score is sufficient to win. Under Hahn scoring, a rational player would probably invade, in order to maximize the expected win. In some sense, "half a point is good enough" may be easier for such situations - the safe strategy is easier to compute; seal the borders and count, if you have enough, you're done. Smart players will economize - "rich men don't pick fights" - the game will progress to simpler, more easily-analyzed paths, where the outcome is certain. In a way, this is like an Indian parable: a sultan decreed that his daughter would be given in marriage to the slowest horse in a race among her suitors. In order to prevent the race from taking all day, he randomly assigned each suitor to ride a different suitor's horse. In regular go, rich men (winners) don't pick fights; losers do. In Hahn go, rich men pick fights, and losers seek to minimize their losses. I'd love to see a regular Hahn tournament among computer programs; it might lead to some interesting advances. Strong programs might become rapaciously bloodthirsty daredevils. They might develop models of opponents' weaknesses - learning that programs A and B always fall for certain swindles, but C and D do not. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
In message <95be1d3b0911241346o3d26135eif8f184eb3f516...@mail.gmail.com>, Vlad Dumitrescu writes On Tue, Nov 24, 2009 at 22:15, Nick Wedd wrote: But the "additive" property of Hahn scoring makes life easy for players. If the board has become separated into regions that do not interact, players can just work out what they think is the biggest local move on each part of the board, and then make the biggest of these moves. This calculation is correct for Hahn scoring, but not for normal scoring, or indeed any other way of scoring. (I am not talking about tedomari effects here, which are rare and small; I am talking about the handling of uncertainty.) In fact, I believe that of all the ways of converting from the board score to the object of the game, Hahn scoring is the uniquely easiest and least interesting. I'm sorry to bother you, but I don't get it. There must be some subtle detail that escapes me... Please try to explain why the "hahn calculation" isn't working in a normal game so as to ensure a win. I'm talking about strong human players. In my view, we have hahn: object of the game = max board score normal: object of the game = board score > komi Are you talking about omniscient players? If not, I have already answered: > Suppose my attempts to read the game tell me "If I seal off my > territory at A, I will win by 5 points. If instead I invade at > B, then 70% of the time I will win by 25 points, 30% of the time > I will lose by 5 points". > If I am playing Go, I will prefer A. If I am playing bang neki, > [or Hahn scoring] I will prefer B. Nick -- Nick Weddn...@maproom.co.uk ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: A cluster version of Zen is running on cgos 19x19
> Also, on 19x19 board, current 16-core cluster version performs almost > the same as 8-core shared memory pc such as Mac Pro, which Yamato used > for KGS. Hi Hideki, Is that difference due to a scaling limit of Zen, or is this due to the cluster overhead? Would moving from gigabit to infiniband help, or is the limit more to do with the lack of shared memory? >T2K HPC cluster This seems to be a cluster specification rather than an actual machine. Can you tell us more about how many cores you are experimenting with, and how the programs scale? (Are all your experiments with Zen, or are you trying to run other programs on a cluster too?) 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: Hahn system tournament and MC bots
No professional gambler, if he had the numbers laid out for him, would
ever choose unoptimal play, not when he's playing for the long
term. The computer, in the same way, would have to be modeled to
maximize expected value. Nothing else makes sense.
In a single game with high stakes, yes mindset (risk aversion, your
finances, etc) might come into it. But that is exactly why I specified
the long term scenario.
-Jeff
A professional gambler has a 2 step task.
1. Find a weaker player (aka "fish")
2. capture the fish('s bankroll)
Luckily for the gambler, there is usually an adequate supply of fish.
But capturing the fish is a bigger problem.
Fish are shy and need to fed worms to make them feel comfortable.
So step 2 is really a 3 step process:
2a. let the fish win at the relatively small starting stakes.
2b. Act angry and astounded, offer a raise of the stakes. Now start winning.
2c. The angry and astounded fish will offer another raise of the stakes if
step 2b was artful enough. Now capture the fish.
So the whole idea of "optimizing" the score it totally besides the point.
Stefan
___
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Wed, Nov 25, 2009 at 12:11:55AM +0100, Stefan Kaitschick wrote: > > A professional gambler has a 2 step task. > 1. Find a weaker player (aka "fish") [...] > So the whole idea of "optimizing" the score it totally besides the point. I was using the professional gambler as a rational player in an idealized and defined setting. The point was to get around all the talk about player modelling, gods and devils, fish, and whatnot. Sure all of that other stuff is interesting and appropriate in a real setting, but the discussion wasn't getting anwhere. Even a professional poker player looking for fish or playing in tournaments understands the basic probabilities and fundamental concepts like implied odds that have nothing to do with psychology or player skill. -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots. and KGS tournament ?
In message <200911242252.09463.alain.baecker...@laposte.net>, Alain Baeckeroot writes In another thread Nick Wedd wrote: The December KGS bot tournament will be 9x9. I guess that if a cluster-Zen competes in that (I am hoping it will), it will be unbeatable. The existing pattern of KGS bot tournaments (see http://www.weddslist.com/kgs/future.html) means that the January one will also be 9x9, then February and March will both be 19x19. ... Is there a possibility for an Hahn tournament on KGS ? maybe with simplified rules: one point on board is one point in tournament ( (c) R.Jasiek ) The tournaments I run on KGS use the server's tournament manager. This makes my job much easier. But it knows nothing about Hahn scoring. Two things I could do: 1.) Run a tournament manually, telling the operators who their opponents in each round will be, and adding up the score myself. I am not very keen on this, I see to much room for error. 2.) Use the tournament manager, and let it plan the pairings based on its own opinion of who is doing well in the tournament (this won't be too far from reality, Aya will beat WeakBot50K either way). But declare the result based on the total Hahn score of the players. I would prefer (2). I would be willing to hold a tournament like that. Nick -- Nick Weddn...@maproom.co.uk ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] OT: gambling (was: Hahn system...)
>> No professional gambler, if he had the numbers laid out for him, would
>> ever choose unoptimal play, ...
>
> A professional gambler has a 2 step task.
> 1. Find a weaker player (aka "fish")
> 2. capture the fish('s bankroll)
Big Deal, by Anthony Holden, is a fine read (a professional writer took
a year off to become a poker pro), and nicely shows the balance between
maths, bluffing and hustling by *professional* gamblers.
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/
[computer-go] Optimizing combinations of flags
>What do you do when you add a new parameter? Do you retain your existing >'history', considering each game to have been played with the value of >the new parameter set to zero? Yes, exactly. >If you have 50 parameters already, doesn't adding a new parameter create >a rather large number of new parameter sets, most of which there will >never be time to investigate? Yes. So the new parameter will drift to its optimal value against the existing parameter values. But here's the thing: declining epsilon greedy policies are zero regret starting from any initial state. So if the setting of the new parameter affects old parameter settings, then the established parameters will start to move as well. If the objective function is a convex function of the parameters (which is generally the case, based on the curves that I have seen) then the whole system will drift to a global optimum. >I have been using UCB and UCT to tune engine settings, but I don't think >these methods work well to tune more than a handful of parameters at the >same time. Such systems have trouble because their exploration is a *deterministic* function of the sequence of wins. That is, all parameters will lock into the same set of feedback. If you use UCT, then you have to optimize *combinations* of parameters, which is unwieldy. Declining epsilon greedy is a randomized exploration strategy, but still zero-regret. Now the same sequence of wins/losses can be used to tune all parameters concurrently. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: A cluster version of Zen is running on cgos 19x19
Darren Cook: <4b0c6706.7070...@dcook.org>: >> Also, on 19x19 board, current 16-core cluster version performs almost >> the same as 8-core shared memory pc such as Mac Pro, which Yamato used >> for KGS. > >Hi Hideki, >Is that difference due to a scaling limit of Zen, or is this due to the >cluster overhead? Would moving from gigabit to infiniband help, or is >the limit more to do with the lack of shared memory? I'm right now evaluating the scaling (:-). The performance gap is perhaps due to the algorithms. Almost all cluster versions of current strong programs (MoGo, MFG, Fuego and Zen) use root parallel while shared memory computers allow us to use thread parallelism, which gives better performance. The main reason, I guess, is that the latter increses the depth of the search tree according to the number of processors (cores) while the former does not. One interesting observed thing of root parallel is that the scaling depends on the time for a move; longer time setting shows better scalability, when the time period to exchange root information is fixed. In other words, each time setting has its best number of nodes. This makes things complicated :(. The scaling limit of Zen is still unknown, though I expected that the playouts of Zen was not so random that it did not scale well, before starting this joint project with Yamato. >>T2K HPC cluster > >This seems to be a cluster specification rather than an actual machine. >Can you tell us more about how many cores you are experimenting with, >and how the programs scale? (Are all your experiments with Zen, or are >you trying to run other programs on a cluster too?) I'm running only Zen on the cluster, though I'd like to run my Fudo Go as well if I have (had?) time. Name: T2K Open Supercomputer (Todai) #Todai is an abbreviation of University of Tokyo in Japenese. Hardware: HITACHI HA8000-tc/RS425 Number of nodes: 952 Number of cores of each node: 16 #I can use up to 64 nodes; 1024 cores in total Processor: AMD Opteron 8356 (quad-core) 2.3 GHz Memory of each node: 32 GB Interconnect: Myricom Myri-10G Operating System: RedHat Enterprise Linux 5 #Flops numbers are omitted. :) http://www.cc.u-tokyo.ac.jp/service/ha8000/intro.html (in Japanese) T2K stands for Tokyo, Tsukuba and Kyoto (T, T, K). See http://www.open-supercomputer.org/ (in English) for the idea of T2K Open Supercomputer. Hideki -- g...@nue.ci.i.u-tokyo.ac.jp (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] Re: [computer-go] Re: A cluster version of Zen is running on cgos 19x19
> > The performance gap is perhaps due to the algorithms. Almost all > cluster versions of current strong programs (MoGo, MFG, Fuego and Zen) > use root parallel while shared memory computers allow us to use thread > parallelism, which gives better performance. I think you should not have troubles with your networks, at least with the number of machines you are considering. Perhaps you should increase a little the time between two communications ? With something like mpi_all_reduce for averaging the statistics over all the tree at each communication, more than 3 or 4 communications per second is useless. Averaging statistics in nodes with less than 5% of the total number of simulations might be useless also. Best regards, Olivier ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
2009/11/24 terry mcintyre : >>Please try to explain why the "hahn calculation" isn't working in a >>normal game so as to ensure a win. I'm talking about strong human >>players. > >>In my view, we have >> hahn: object of the game = max board score >> normal: object of the game = board score > komi >>Both seem just as easy and interesting. > > If you are winning in the Hahn sense, your score also exceeds komi; but Hahn > scoring - either by accumulating points in a tournament ranking, or > converting points to dollars in bang neki fashion, gives you incentive to > achieve larger scores. Thank you,Terry, for the explanation. I understand (and I think I already said it too) that normal go is easier. What Nick said before was: "This calculation is correct for Hahn scoring, but not for normal scoring, or indeed any other way of scoring." The way I read it, just because one way is easier doesn't mean it is incorrect in other cases. My question was about why he thinks the calculation is incorrect for other ways of scoring. best regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Hahn system tournament and MC bots
On Tue, Nov 24, 2009 at 23:58, Nick Wedd wrote: > Vlad Dumitrescu writes >> Please try to explain why the "hahn calculation" isn't working in a >> normal game so as to ensure a win. I'm talking about strong human >> players. > > Are you talking about omniscient players? If not, I have already answered: > >> Suppose my attempts to read the game tell me "If I seal off my >> territory at A, I will win by 5 points. If instead I invade at >> B, then 70% of the time I will win by 25 points, 30% of the time >> I will lose by 5 points". > >> If I am playing Go, I will prefer A. If I am playing bang neki, >> [or Hahn scoring] I will prefer B. Hi Nick, I'm not talking about omniscient players. My problem with your answer above is that "70% of the time I will win by 25 points" contains implicit assumptions, which affect the exact numbers and thus the choice one would make. How is the 70% calculated? Against whom? The ratio will be different against someone 3 stones stronger, against someone of my strength or someone three stones weaker. Do the playlines leading to a 25 points win involve a stupid mistake that the opponent must make, or are they about luring him gently in a non-optimal position? Or is it a big fight that is involved, I can't read it but feel 70% confident I can win it? Opponent modeling will also play a role and then the percentage can't even be computed anymore, the player has to guess, to gamble. It is here that one's personality comes to play. "This guy doesn't like to invade deep, let's expand this moyo". "That guy is a fierce fighter, better to play it safe". Personally, I would almost always play A in a tournament situation (or with money involved). In a Hahn tournament, it depends on the overall situation, if a small win in this game isn't enough then I must try B if I want to win. If it's for fun, then exploring B is much more interesting and offers more opportunity to learning. best regards, Vlad ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [SPAM] Re: [computer-go] Re: A cluster version of Zen is running on cgos 19x19
Thank you Oliver, Olivier Teytaud: : >> >> The performance gap is perhaps due to the algorithms. Almost all >> cluster versions of current strong programs (MoGo, MFG, Fuego and Zen) >> use root parallel while shared memory computers allow us to use thread >> parallelism, which gives better performance. > > >I think you should not have troubles with your networks, at least with >the number of machines you are considering. > >Perhaps you should increase a little the time between two communications ? >With something like mpi_all_reduce for averaging the statistics over all the >tree at each communication, more than 3 or 4 communications per second >is useless. Averaging statistics in nodes with less than 5% of the total >number of simulations might be useless also. In your (or Sylvain's?) recent paper, you wrote less than one second interval was useless. I've observed similar. I'm now evaluating the performance with 0.2, 0.4, 1 and 4 second intervals for 5 second per move setting on 19x19 board on 32 nodes of HA8000 cluster. Though I have not enough games yet, current best is 1 second interval which improves about 400 Elo in self-play. Then, why we have similar experiments with different implementations of root parallelism, based on different programs and on different clusters? I don't use MPI for the cluster version of Zen. Zen's playouts are slower than MoGo's. Etc... One second is a mysterious time :(. Hideki -- g...@nue.ci.i.u-tokyo.ac.jp (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
