[computer-go] Re: Dynamic Komi's basics
Darren Cook wrote: This (if I've understood correctly) is what I thought the dynamic komi idea was, i.e.: Aim to be winning 60% of simulations. If winrate is over 60, increase the artificial komi (if black; decrease it if white) on the next move (*). If winrate is below 40, then do the opposite. In my original experiments (in Summer 2009) I let the bot use dynamic komi when it was stronger than the opponent (and thus had to give handicap): RULE 42 I started with a reasonable komi-bonus for the bot (leading to a 30+ percentage for the bot) and kept this until the bot climbed above 42 %. Then the komi value was reduced by 10 or 8 or 5 points. The procedure worked well for instance when MC-Many Faces played against a non-MC version of MF. This should work well on all board sizes, for both black and white, and in all stages of the game (though increasing the 60% threshold as the game progresses may make sense). Right. I'm surprised people are using a simple linear decreasing rule, but very interested to hear there is a tangible improvement. I was surprised, too. Perhaps being adaptive isn't needed? The simplest solution is the best solution (old proverb). 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/
Re: [computer-go] Re: Dynamic Komi's basics
Quoting Ingo Althöfer 3-hirn-ver...@gmx.de: Darren Cook wrote: I'm surprised people are using a simple linear decreasing rule, but very interested to hear there is a tangible improvement. I was surprised, too. Perhaps being adaptive isn't needed? The simplest solution is the best solution (old proverb). I have implemented *adaptive* dynamic komi in Valkyria now and are testing it on KGS. It adjusts the komi after each move, and uses the same komi for each generated move. Wild stuff do happen when groups dies and both colors makes blunders under time pressure. With adaptive dynamic komi the search becomes very unpredictable. So for many moves it will search with a too high or a too low komi. In the opening it is not much of a trouble. So I can perfectly see why a linearly decreasing komi will give a boost for sure to the playing strength without risking loosing games because of the chaotic nature of adaptive dynamic komi. Valkyria now uses dynamic komi in all games and also for lost position up to being behind 20 points. It sets the dynamic part to zero when there are two dames left, which leads to the tricky behavior that it tries wild things just before resigning, after trying to catch up by playing a proper endgame. Tonight it move from 6 kyu to 5 kyu, but the 6 kyu rating was not certain so it is hard to tell whether the change affects the strength. I think it is now more fun to play against it and would appreciate any comments about the new playing style from strong as well from weak players. Best Magnus -- Magnus Persson Berlin, Germany ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] CG2010 extended deadline
Hi, In case you are submitting a paper to the conference in Kanazawa, you may like to know that the deadline has been extended to 2010-02-25: https://www.conftool.net/cg2010/index.php?page=index Here is the call for papers: http://www.jaist.ac.jp/ICGA-events-2010/english/conference/ Rémi ---BeginMessage--- Dear Remi, Can you announce that the deadline for the cfp is changed into February 25, 2010. Thank you, Joke ---End Message--- ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Strong programs on cgos 19x19?
or the strong version of pachi. Done. Jean-loup 2010/2/16 David Fotland fotl...@smart-games.com My old MPI code had a scaling bug. Performance scaling (playouts per second) was linear, but the strength did not scale well, and 64 cores was weaker than 32 cores. I have a 16 core cluster of my own now (four 2.3 GHz Q8200 quad core), and I discovered that the MPI code hangs when using MPICH2 rather than the Microsoft MPI library. So last week I rewrote it with a different algorithm, and it seems to scale much better. It's on CGOS now, for at least a few days. It would be great if some other strong programs could join, zen, or fuego-8c, or aya-4c, or the strong version or pachi. Thanks, David ___ 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] Dynamic Komi at 9x9 ?
Hello, I informed the German go scene that there is (some) progress at KGS bots with dynamic komi. Based on this, a friend told me that they would have an open afternoon for go beginners in the middle of March - and they expect many newbies with strengths between 17k and 30k. His question is if a bot with dynamic komi might be a suitable opponent for such beginners on 9x9 (with high handicap). Does someone here have already experience with non-static komi for handicap games on 9x9? Or would someone be willing to test something with his bot? Ingo. -- NEU: Mit GMX DSL über 1000,- ¿ sparen! http://portal.gmx.net/de/go/dsl02 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Dynamic Komi at 9x9 ?
Fuego_9x9_1h (or a variation of this name) played on OGS a couple of handicap 9x9 games. It used dynamic komi. I think it was manually adjusted after every move, and worked well. -ibd Am 17.02.2010 um 22:51 schrieb Ingo Althöfer: Hello, I informed the German go scene that there is (some) progress at KGS bots with dynamic komi. Based on this, a friend told me that they would have an open afternoon for go beginners in the middle of March - and they expect many newbies with strengths between 17k and 30k. His question is if a bot with dynamic komi might be a suitable opponent for such beginners on 9x9 (with high handicap). Does someone here have already experience with non-static komi for handicap games on 9x9? Or would someone be willing to test something with his bot? Ingo. -- NEU: Mit GMX DSL über 1000,- ¿ sparen! http://portal.gmx.net/de/go/dsl02 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Dynamic Komi at 9x9 ?
Does anyone have data based on several thousands games that attempts to measure the effect of dynamic komi?I would like to see results that are statistically meaningful. We need to see a few thousand games played against a fixed opponent WITH dynamic komi, and then the same program without dyanmic komi playing against the same opponent with the same number of games.The number of games must be decided before the test is run, or the error margin calculation is meaningless. As far as I can tell, nobody has yet to produce anything more than anecdotal evidence that this works. Having a person manually adjusting this after every game is completely non-sceientific, unless they are doing it in a fixed way with no decision making on their part and they are playing thousands of games (or at least enough to get statistically significant results.) I'm not trying to rain on anyone's parade, but I cannot understand why no one has produced a statistically meaningful result on this subject - or if I'm wrong please point me to the paper or data and games that were played. I am genuinely interested in this since I never was able to make it work when I spent about one intense week on it.(I did not do this with handicap games, but with normal games.) Don On Wed, Feb 17, 2010 at 4:57 PM, Isaac Deutsch i...@gmx.ch wrote: Fuego_9x9_1h (or a variation of this name) played on OGS a couple of handicap 9x9 games. It used dynamic komi. I think it was manually adjusted after every move, and worked well. -ibd Am 17.02.2010 um 22:51 schrieb Ingo Althöfer: Hello, I informed the German go scene that there is (some) progress at KGS bots with dynamic komi. Based on this, a friend told me that they would have an open afternoon for go beginners in the middle of March - and they expect many newbies with strengths between 17k and 30k. His question is if a bot with dynamic komi might be a suitable opponent for such beginners on 9x9 (with high handicap). Does someone here have already experience with non-static komi for handicap games on 9x9? Or would someone be willing to test something with his bot? Ingo. -- NEU: Mit GMX DSL über 1000,- ¿ sparen! http://portal.gmx.net/de/go/dsl02 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Strong programs on cgos 19x19?
The strong pachi is really strong! What hardware is it running on? Can you say how it differs from the vanilla pachi? David From: computer-go-boun...@computer-go.org [mailto:computer-go-boun...@computer-go.org] On Behalf Of Jean-loup Gailly Sent: Wednesday, February 17, 2010 9:50 AM To: computer-go Subject: Re: [computer-go] Strong programs on cgos 19x19? or the strong version of pachi. Done. Jean-loup 2010/2/16 David Fotland fotl...@smart-games.com My old MPI code had a scaling bug. Performance scaling (playouts per second) was linear, but the strength did not scale well, and 64 cores was weaker than 32 cores. I have a 16 core cluster of my own now (four 2.3 GHz Q8200 quad core), and I discovered that the MPI code hangs when using MPICH2 rather than the Microsoft MPI library. So last week I rewrote it with a different algorithm, and it seems to scale much better. It's on CGOS now, for at least a few days. It would be great if some other strong programs could join, zen, or fuego-8c, or aya-4c, or the strong version or pachi. Thanks, David ___ 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/