Re: [Computer-go] Aya won December 2014 KGS bot tournament
http://dvandva.org/pipermail/computer-go/ On 12/08/2014 10:40 PM, Hiroshi Yamashita wrote: By the way, I can not see archives now. http://computer-go.org/pipermail/computer-go/ ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] 8th UEC Cup + 3rd Densei-sen
Hi, I have just noticed: UEC Cup on March 14-15: http://jsb.cs.uec.ac.jp/~igo/eng/ Densei-sen on March 17th, with Cho Chikun: http://entcog.c.ooco.jp/entcog/densei/ http://entcog.c.ooco.jp/entcog/densei/share/data/EC_press_densei.pdf Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to AyaMC!
Thanks Hiroshi. Most of the paper is available on google books: http://books.google.fr/books?id=52kqBAAAQBAJpg=PA26#v=onepageqf=false Rémi On 22/11/2014 18:53, Hiroshi Yamashita wrote: Delayed congratulations to AyaMC, winner of last Sunday's KGS bot Thank you for the tournament and report, Nick! Recently I got +100 Elo from selfplay by adding static bonus in UCB. I used Ikeda's paper method. Effciency of Static Knowledge Bias in Monte-Carlo Tree Search Kokolo Ikeda and Simon Viennot, CG2013 UCB is like this. UCB = w/n + C * sqrt( log(N) / n ); RAVE = Rw/Rn + C * sqrt( log(N*175) / (N*0.48) ); beta = Rn / (Rn + n * (W1 + W2 * Rn)); UCB_RAVE = beta*RAVE + (1-beta)*UCB + G*log(1+gamma)*sqrt( K / (K + n)); n : child nodes w : child wins Rn: child nodes (Rave) Rw: child wins (Rave) N : sum of children's nodes gamma : child gamma from MM C = 0.31 W1 = (1.0 / 0.9) W2 = (1.0 / 2) K = 600 G = 0.01 Regards, Hiroshi Yamashita ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Zombie processes
The problem is I was using Windows, and I am not very familiar with that system. In Linux, the child would have died automatically. Rémi On 27/10/2014 10:08, Petr Baudis wrote: Hi! On Sun, Oct 26, 2014 at 10:30:10AM +, Nick Wedd wrote: I wonder how practicable it would be for a bot, on startup, to search for other copies of itself running, and issue a warning if it finds any. I think it would be much easier in practice if the program made sure it does not keep running when its parent process (kgsGtp) dies. This can happen in multiple ways - either by doing this explicitly, http://stackoverflow.com/questions/284325/how-to-make-child-process-die-after-parent-exits has a couple of ideas, but even a much more elegant way is simply not overriding SIGPIPE. Then your process will get that signal when trying to reply to a deceased kgsGtp and die automatically. I suspect these problems stem from disabling SIGPIPE which really shouldn't be necessary for anything in this context anyway. Petr Baudis ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] codecentric Challenge
Thanks Ingo. I enjoyed this match very much. Rémi On 27/10/2014 11:13, Ingo Althöfer wrote: Hello, it seems that currently the mails from this list are not properly distributed, at least I did not receive the two mails on zombie processes. Nevertheless, I found them in the archive. Here are some replies: Nick Wedd wrote: Yesterday F J Dickhut won his game in the codecentric go challenge 2014 five-game match against Crazy Stone, bringing the score to 3-1, and winning the match. My congratulations to Franz Dickhut! Congratulations also from me to Franz-Josef. He fully deserves the win (and the prize money). It was interesting to see his learning process: From round to round he had CrazyStone better under control. I hope such an event will be held again next year. It is my intention to organize a similar event in Fall 2015, with a top German amateur player and a top computer program. On the zombie processes: It is sort of a miracle for me that Remi did not realize the double load quickly. Normally the bot is showing its simulation count all the time, and smaller numbers should be recognized more or less immediately. I like Nick's proposal for process self control. At least it should be an option. At other occasions I perform experiments with CS vs CS on one machine, and there it would be a catastrophe when one process would shut down his brothers. *** Olivier wrote: By the way even winning just one game is a great success (no handicap!). Indeed it is. And hopefully the event and my plan to have another one in Fall 2015 will motivate programmers to try for success. One of things Remi might try to implement should be knowledge about the 1000-year ko (that happened in round 2 in the upper right corner). Cheers, Ingo. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Interviews on codecentric Challenge
Hi, As promised, here is the analysis of the first game by Crazy Stone: http://remi.coulom.free.fr/CrazyStone/analysis/2014-10-04-fj-CrazyStone/index.html sgf: http://files.gokgs.com/games/2014/10/4/fj-CrazyStone.sgf Thanks FJ, thanks Ingo. I am looking forward to the next game. Rémi On 10/2/2014 5:22 PM, Ingo Althöfer wrote: Hi Nick, thanks for the help. So, I wonder if you can tell me the account names to be used by the players, fj will be Franz-Josef Dickhut CrazyStone will be CrazyStone and what room the games will be played in? The Computer Room. Cheers, Ingo. PS. Please, do not wonder about codecentric with a small c in front. That is official policy of the company. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Interviews on codecentric Challenge
Thanks. It was not a laptop, but a big desktop. i7-5930K, 3.5 GHz, 6 cores, 12 threads. I left the university one month ago, so I don't have access to the 24-core server any more. Those 6 cores are really fast. About 3.3 k playouts/s on one core, compared to about 2k playouts/s on one core of the 24-core AMD (only playout, no tree search). And Hyperthreading is rather efficient. With tree search, I get about 15.3k playouts per second with 6 threads, and 25.5k playouts per second with 12 threads. All those numbers are from the empty starting position. Rémi On 10/5/2014 7:09 PM, Xavier Combelle wrote: Concratulation according the fact that it was just the laptop power http://www.talkchess.com/forum/viewtopic.php?topic_view=threadsp=591192t=53262 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] some general UCT notes.
On 30 août 2014, at 22:10, Dave Dyer dd...@real-me.net wrote: I've also been comparing blitz play which creates a copy of the board at top level, and starts each descent with a copy of the board; compared with unwinding play where every move is explicitly unwound. Of course, the complexity of the unwinding varies a lot from game to game, but I found that unwind is always faster, an average 1/3 faster across several games. So if the complexity of unwinding your data structures is not too great, it's worthwhile. Hi Dave, I am very surprised by this. It is certainly not the case for Go, even on 9x9. Making a copy of the board should be orders of magnitude faster than running a playout. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] ICGA Journal
I am not a member of the ICGA either. I think it is really ridiculous to have a paper-only computer-science journal in 2014. I’d like to suggest to Detlef to make his paper available online somewhere and post a link to this list. I am looking forward to reading it. Rémi On 21 août 2014, at 14:17, Petr Baudis pa...@ucw.cz wrote: Hi! On Thu, Aug 21, 2014 at 12:44:19PM +0200, Ingo Althöfer wrote: - at least I hope it's still being published, ??? (So you are not a member of the ICGA - otherwise you would know better) Unfortunately not right now, as my Computer Go research profile is very low in recent years. Of course, the ICGA Journal keeps being published. I have all issues until end of 2013. And the March 2014 issue will come soon (delay of 3-5 months is an old ICCA/ICGA Journal tradition). Then I'd suggest that the homepage is updated so that it doesn't list 34-1 from March 2011 as the latest issue (both as direct link and in the contents database). I fear it might be turning some potential authors away. Kind regards, Petr Baudis ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Update on the ICGA Journal
I can’t get the TOC there either. And it has not been indexed by DBLP since 2011: http://www.informatik.uni-trier.de/~ley/db/journals/icga/index.html IEEE TCIAIG makes a much better impression: http://www.informatik.uni-trier.de/~ley/db/journals/tciaig/index.html http://ieeexplore.ieee.org/xpl/RecentIssue.jsp?punumber=4804728 You can also compare their sjr ratings: http://www.scimagojr.com/journalsearch.php?q=19700177025tip=sidclean=0 http://www.scimagojr.com/journalsearch.php?q=4500151521tip=sidclean=0 I don’t know how many copies of the ICGA Journal are distributed. I expect around 100 or 200. And a very large part of the subscribers are not interested in Go at all. You’d make a much bigger impact by publishing your idea on this list. Rémi On 21 août 2014, at 19:02, Petr Baudis pa...@ucw.cz wrote: Hi! On Thu, Aug 21, 2014 at 02:46:11PM +0200, Ingo Althöfer wrote: Then I'd suggest that the homepage is updated so that it doesn't list 34-1 from March 2011 as the latest issue (both as direct link and in the contents database). It seems you have been looking on an outdated website. From this site http://icga.uvt.nl/?page_id=26#blank you get at least the TOCs until June 2013. Frankly, I can't figure it out. :-( Maybe it should be available when I click TOC in the Navigation section, but literally nothing happens at that point, the page doesn't change or anything. (Both in Firefox and Chrome.) (Still, I think my point stands - ICGA journal google query will bring the original page http://ticc.uvt.nl/icga/journal/ with old Latest Issue as the first result to me and nothing on it indicates to me that I should look further.) Kind regards, Petr Baudis ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] [ANN] Imago - Go board optical recognition
On 13 août 2014, at 10:35, Petr Baudis pa...@ucw.cz wrote: Hi! On Tue, Aug 12, 2014 at 01:15:11PM +0100, Aja Huang wrote: Several cameras are relaying the games of the best players. A smart optical recognition program automatically converts the streaming images to sgfs and sends them to a Go program. The Go program then shows rich analyses over the games, such as wining rate, best moves, principal variations, estimated territories, etc, even predicts the next moves. A spectator is watching a friend's game and wondering who is ahead. He doesn't understand Go very well. He uses his android phone takes a snapshot. After 3 seconds, the Go software oh his phone immediately tells him that his friend is ahead for 20 points and winning with 90% chance. I'm sure we will get there, and maybe it won't take so long. :-) Working on a reliable recognition could be a good first step. I think analysis of the top boards is the most difficult part as it's similar difficulty as playing on that level. I think Remi is best positioned to build something like this as he has most of the pieces. Otherwise, we'd need a sort of (ideally open) ecosystem in the mobile, mainly an app that brings the engine, analysis visualization, recognition and some Go GUI all together. Doesn't require any scientific breakthroughs, just patience and technical skill. There was a bit of irony in Aja’s post, because kifu-snap Crazy Stone already did this in Sibiu. That’s the reason for the last sentence of his message (the one you did not quote). I know the monthly subscription system that Unbalance used for this app turned out extremely unpopular, but they are not willing to change it now. Note that you can try the app for free for 10 days. If you don’t want to pay, then just be careful to cancel your subscription before the end of the 10 days. I am confident the recognition rate of kifu snap could be improved a lot. But I’d rather not claim too much, because it would not be the first time I underestimate the difficulty of such tasks. It is a really exciting problem. I’ll certainly work on it again when I have some free time, and I’ll follow your progress with a lot of interest. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] [ANN] Imago - Go board optical recognition
Thanks Petr, Thanks Thomas, Very interesting report. I do something similar in kifu-snap. Instead of RANSAC, I use another Hough Transform to detect the sinusoids in the first Hough Transform. I also use k-means color clustering for detecting stones, but also combine this with other features (like circle detection). So kifu-snap should work with black-and-white diagrams, for instance. I used a lot of tricks to make my implementation fast (because it has to be fast on a phone), so it takes only 0.1 second on average to process one picture on one core of my 2.4 GHz core2 duo notebook. Your dataset is way too easy, though. I tried only a few pics, and kifu-snap had no difficulty with them. I uploaded my own dataset there: http://remi.coulom.free.fr/kifu-snap/goban.tar.bz2 I don’t own the rights for most of these photos. Found most of them with Google images. If anybody complains, I will remove the file. The .ks files are coordinates of the corners of the board (pixels, with 0,0 at the center of the image, IIRC). The .gob files are stone configurations. This is the result of stone-recognition (given the correct grid placement from the .ks file): http://remi.coulom.free.fr/kifu-snap/stone-recognition.txt (time is in milliseconds) Only 32/99 grids placements are recognized correctly. Most frequent mistake is to be off by one or two lines in one direction. Only 14/99 are fully recognized automatically. I admit some of them are particularly tricky. But I also have plenty of pictures from real usage of kifu-snap, and it is still difficult to be correct. When I use kifu-snap at the Go club or in a tournament, I usually get a 100% automatically recognized board a bit more than 50% of the time (completely biased guessed rate). Automatic Go board recognition is a really difficult and fun challenge. Rémi On 12 août 2014, at 13:35, Petr Baudis pa...@ucw.cz wrote: Hi! Tomas Musil (a student of mine), has created a state-of-the-art open source Go board optical recognition software. We have focused on completely automatic runs, so it automatically detects the board corners and then the stones on the board, and the precision seems pretty good at least in reasonable lighting conditions. You can find it at http://tomasm.cz/imago together with a lot of pictures, documentation and bachelor thesis describing the algorithms in detail. In the thesis, Tomas also compares it against other similar apps, and it appears Imago shows the best performance of all these that were available to us. Unfortunately, we specifically couldn't easily compare it to Remi Coulom's Kifu-snap for multiple reasons - mainly because that is a mobile app. Hopefully, someone will be able to compare these two in the future. At any rate, I think Imago is a great starting point for anyone who would like to play with Go board recognition. My personal dream would be if we added video capability and further improved speed + reliability in time for EGC2015 (in Czech Republic) and were able to deploy it there to transfer large number of top boards. But this will depend on how much time Tomas will have after the summer (and we didn't actually check with EGC2015 organizers yet), so it's still more of just a dream. :-) Petr Baudis ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Skip-opening matchmode
On 5 juin 2014, at 14:54, Stefan Kaitschick stefan.kaitsch...@hamburg.de wrote: such as Remi Couloms simulation balancing Simulation Balancing is not mine. I did not even implement it ;-) Rémi___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Bitwise-parallel surround capture
Hi Cameron, I did not take time to read your paper in details, but I know one go program based on bitboards: http://www.quirkster.com/iano/forth/fgp.html I don’t know if it manages captures differently from you, but you might at least wish to cite it. Rémi On 24 mai 2014, at 11:44, Cameron Browne cameron.bro...@btinternet.com wrote: Hi, This draft paper describes a simple bitwise-parallel method for performing surround capture: http://www.cameronius.com/research/go-bits-1.pdf Before I submit it, I just wanted to check with this list that the method is not known and already in use. Any general comments on the paper would also be appreciated. Regards, Cameron ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Bitwise-parallel surround capture
I found the code there: http://comments.gmane.org/gmane.games.devel.go/20854 On 24 mai 2014, at 18:28, René van de Veerdonk rene.vandeveerd...@gmail.com wrote: On Sat, May 24, 2014 at 4:34 AM, Erik van der Werf erikvanderw...@gmail.com wrote: I think a speed comparison for well optimized code using pure bit-boards (no liberties) against using pseudo-liberties (e.g., as in early versions of Lukasz Lew's libEGO) would be of some interest. I did just that a few years back and got to similar performance for 9x9, whereas bitmaps were slower at 19x19. Test-code and results can be found in the archives looking for bitmapgo or alike. If the test-code is missing, I can mail it out to those interested. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Computer games reviewed by Narumi Osawa
Hi, Some of you might be interested in this youtube video, with English translation: https://www.youtube.com/watch?v=gsHK-btcpPM Thanks to the person who gave the link to me. The reviewed games are Zen vs Ishida, and Crazy Stone vs Ishida. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] UEC 2014 official final results?
The final results are not available in English, but they are in Japanese: http://jsb.cs.uec.ac.jp/~igo/result2.html Rémi On 7 avr. 2014, at 19:18, Detlef Schmicker d...@physik.de wrote: Hi, I can not find an official final result page of this years UEC cup. Would be great, as I try to get a hardware sponsor and like to prove, that the world best programs are playing in KGS tournaments... Or is there another good source to prove this? Thanks Detlef ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Game records
Hi Myriam All the KGS game records have time stamps. And they are in sgf format (using BL and WL properties). http://www.red-bean.com/sgf/properties.html#BL Rémi On 14 févr. 2014, at 15:32, Myriam Abramson mabram...@gmail.com wrote: Does anybody know where I could get game records with the timestamp of every move? I know that the sgf format does not support that. It does not have to be about the game of go necessarily. Thanks for any info. -- myriam Go Proverb: Fill in a semiai from the outside. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] 1st Go Denou-sen
It seems Zen lost the 4 games: http://news.mynavi.jp/news/2014/02/11/106/ Rémi On 3 févr. 2014, at 11:01, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: Hi, 1st Go Denou-sen will be held February 11th and 16th. Zen will play against two pros in 9x9 and one top amateur in 13x13, and Ozawa Ichiro in 19x19. Ozawa is a one of the famous Japanese politician. He is known as strong Go player. Games will be broadcasted on web, niconico-video site only. There is no KGS live. February 11th 9x9, 3 games each, no handicaps. komi 6.5, Japanese rule, 20min+30sec. Chang Li Yu, 8d Pro - Zen Hirata Tomoya, 3d Pro - Zen February 16th 13x13, 3 games, no handicaps. komi 6.5, Japanese rule, 30min+30sec. Emura Kikou, 7d Ama - Zen 19x19, 1 game, no handicaps. komi 6.5, Japanese rule, 60min+60sec. Ozawa Ichiro, - Zen Note: This event has no relationship about UEC cup and Densei-sen competision. Go Denou-sen is originated from human-computer event of Japanese chess, Shogi. And it is organized by Dowango Co., Ltd. 1st Go Denou-sen (in Japanese) http://ex.nicovideo.jp/denou/igo/ Promotion Video (in Japanese) http://www.nicovideo.jp/watch/1391105139 Regards, Hiroshi Yamashita ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] 1st Go Denou-sen
Hi Hideki, Thanks for your comments. Are the game records available? Rémi On 11 févr. 2014, at 22:19, Hideki Kato hideki_ka...@ybb.ne.jp wrote: Professinals can play with no or few mistakes on 9x9 but MC bots play with more mistakes (perhaps one per 10 to 20 moves?) due to the randomness. At the two-years-ago event, Zen vs 3 pros, both did many mistakes and Zen had several chances to win. At this event, however, Cho 8p played almost perfectly because he studied 9x9 Go so long (more than 10 hours) and Zen had no chances. Hirata 3p (younger pro) did few mistakes at the 1st game but Zen couldn't take the chance because it was a very narrow sequence of moves. As a conclusion, no-miss-play is required to beat serious professionals on 9x9 :). Hideki Hiroshi Yamashita: 6582F7E197A74985AC2DEB659A9F7753@i3540: It seems Zen lost the 4 games: Yes, Zen lost 4 games in 9x9. Chang Li Yu, 8d Pro - Zen 2-0 Hirata Tomoya, 3d Pro - Zen 2-0 Here is a movie news. Go Denou-sen, pro win decisively (in Japanese) http://www3.nhk.or.jp/news/html/20140211/k10015174611000.html Hiroshi Yamashita ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Hideki Kato mailto:hideki_ka...@ybb.ne.jp ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Elo vs CLOP
MM and CLOP are completely different from each other. MM is for supervised learning of a playing policy from game records. It will tune parameters (pattern weights) to match a given set of samples moves (typically from a collection of game records of strong players). CLOP is black-box optimization, that is to say it can optimize the win rate of your program by letting your program play games against a reference opponent. It will try plenty of different parameter values, play games with them, and try to estimate the best value for the parameter from that. They both optimize something: MM optimizes how the playing policy matches some given sample moves, and CLOP optimizes the win rate against a reference opponent. But you cannot use one method to do the job of the other. Rémi On 11 févr. 2014, at 20:42, Peter Drake dr...@lclark.edu wrote: A naive question: In what situations is it better to use Coulom's Elo method vs his CLOP method for setting parameters? It seems they are both techniques for optimizing a high-dimensional, noisy function. -- Peter Drake https://sites.google.com/a/lclark.edu/drake/ ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to CrazyStone!
Thanks Nick. In round 18, Crazy Stone lost to Zen, not pachi. Now the seki errors of CS vs Zen are already fixed ;-) In the annual table, nomitan gets a “0” but was not in the tournament. The performance of DolBaram is really impressive, considering that it was running on 4 cores. If DolBaram can get good hardware for the UEC Cup, I believe it will have very good chances of winning. Rémi On 4 févr. 2014, at 12:48, Nick Wedd n...@maproom.co.uk wrote: Congratulations to CrazyStone, winner of the February KGS 9x9 bot tournament! My report is at http://www.weddslist.com/kgs/past/100/index.html As usual I look forward to receiving your corrections and comments. Nick -- Nick Wedd n...@maproom.co.uk ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to CrazyStone!
Does Zen evaluate this seki correctly in the playouts? My impression is that it may have passed because it prefers to keep some false hope of winning, rather than be certain of a jigo. Rémi On 4 févr. 2014, at 22:08, Aja Huang ajahu...@gmail.com wrote: Thanks Nick for the report. I'm impressed by DolBaram's performance, in particularly it was using a weaker hardware and scored 1 win, 1 loss and 3 draws against Zen. As you pointed out, seems Zen has a bug on Chinese scoring http://eidogo.com/#url:http://files.gokgs.com/games/2014/2/2/DolBaram-Zen19S-3.sgf Zen (Black) could easily make it a jigo by playing at any point of the the seki area (but W can't) before passing. Aja 2014-02-04 Nick Wedd n...@maproom.co.uk: Congratulations to CrazyStone, winner of the February KGS 9x9 bot tournament! My report is at http://www.weddslist.com/kgs/past/100/index.html As usual I look forward to receiving your corrections and comments. Nick -- Nick Wedd n...@maproom.co.uk ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] 1st Go Denou-sen
Thanks Hiroshi. Nico’s logo is very strange: http://res.nimg.jp/img/base/head/icon/nico/946.gif What does it represent? Rémi On 3 févr. 2014, at 11:01, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: Hi, 1st Go Denou-sen will be held February 11th and 16th. Zen will play against two pros in 9x9 and one top amateur in 13x13, and Ozawa Ichiro in 19x19. Ozawa is a one of the famous Japanese politician. He is known as strong Go player. Games will be broadcasted on web, niconico-video site only. There is no KGS live. February 11th 9x9, 3 games each, no handicaps. komi 6.5, Japanese rule, 20min+30sec. Chang Li Yu, 8d Pro - Zen Hirata Tomoya, 3d Pro - Zen February 16th 13x13, 3 games, no handicaps. komi 6.5, Japanese rule, 30min+30sec. Emura Kikou, 7d Ama - Zen 19x19, 1 game, no handicaps. komi 6.5, Japanese rule, 60min+60sec. Ozawa Ichiro, - Zen Note: This event has no relationship about UEC cup and Densei-sen competision. Go Denou-sen is originated from human-computer event of Japanese chess, Shogi. And it is organized by Dowango Co., Ltd. 1st Go Denou-sen (in Japanese) http://ex.nicovideo.jp/denou/igo/ Promotion Video (in Japanese) http://www.nicovideo.jp/watch/1391105139 Regards, Hiroshi Yamashita ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] KGS bot tournament, this Sunday
Oh, that is the 100th KGS go tournament! Congratulations Nick, and big thanks for your work in organizing them. I appreciate them a lot. Rémi On 31 janv. 2014, at 00:21, Nick Wedd n...@maproom.co.uk wrote: The February KGS bot tournament will be held this Sunday, February 12th, starting at 16:00 UTC and ending at 23:00 UTC. I apologise for the very short notice. It will have 21 rounds, Swiss, with 9x9 boards. The time limits will be 9 minutes each plus very fast Canadian overtime of 10 moves in 30 seconds. The komi will be 7. Note that the integer komi allows jigo as a possible result; if your bot fails to understand this, it may get a jigo where it thought it had a half-point win, or accept a 1-point loss where it could have had a jigo. There are details at http://www.gokgs.com/tournInfo.jsp?id=874 . Please register by emailing me, with the words KGS Tournament Registration in the email title, at mapr...@gmail.com . Nick -- Nick Wedd n...@maproom.co.uk ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] KGS bot tournament, this Sunday
I believe botnoid was MC. On 31 janv. 2014, at 11:15, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: Oh, that is the 100th KGS go tournament! Awesome! 1st KGS tournament was held in 2005, 10 years ago. And GNU Go won, there were no Monte Carlo programs. http://www.weddslist.com/kgs/past/1/index.html I'm curious what kind of program will win in next 10 years. Hiroshi Yamashita ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Anybody going to EGC 2014?
Hi, I am considering the possibility to attend the EGC in Sibiu this summer: http://egc2014.com/ Two weeks is too long for me. I am not familiar with the tournament, but I suppose it is possible to play just a few rounds of the main tournament. And I’ll probably try to participate in a short event, like the 9x9 tournament. There is no computer-go event planned, but if some programmers attend, we can meet and maybe play a few games. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Browsing the game tree?
In Crazy Stone I have a gogui-analyze command that will dump the node of the tree for the current position. So I can just navigate with gogui and ask for the statistics of the current node. It is a bit primitive, but I find it usable enough, and the coding effort is very small. Guillaume Chaslot made a gogui extension long ago, but it seems it disappeared from the web: http://www.mail-archive.com/computer-go@computer-go.org/msg06425.html Rémi On 22 déc. 2013, at 20:47, Petr Baudis pa...@ucw.cz wrote: Hi! I'm wondering how are you investigating your MC game trees? For example, I'm interested in why Pachi did not consider move X, in what branches it _was_ considered and how did the simulations that considered it end up, what are its RAVE stats, etc. So far, I have been relying on debug prints and text dumps of the tree, but they are getting very inconvenient to use for long thinking times and deep trees. My worst case scenario is developing some data format and writing a few tools for visualization and investigation of game trees, but I'd be more than happy to find out that something I could use already exists. Thanks, Petr Pasky Baudis ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Don Dailey
I am sad to report very bad news about the health of Don Dailey. You can read the messages of Larry Kaufman there: http://www.talkchess.com/forum/viewtopic.php?t=50114 Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Playouts vs playingstrength
100 pts/rank is really very wrong. I don’t have data here, but I remember when I ran WHR experiments on KGS data, it was very clear that Elo points / rank increases a lot with rank. At the strength of Crazy Stone (KGS 5d), a rank is worth much more than 200 Elo points. For beginners, it may be somewhere around 50 Elo points. Elo points/handicap stone probably approaches infinity when getting close to optimal play. Rémi On 20 nov. 2013, at 12:17, Petri Pitkanen petri.t.pitka...@gmail.com wrote: EGF uses GOR which to my understanding is based on Elo model. With arbitrary equality to old systems 2100 GOR = 1 dan and 100 pts/rank 2013/11/20 Nick Wedd n...@maproom.co.uk On 20/11/2013 06:45, Detlef Schmicker wrote: After turning off pondering in pachi the plot looks much more sensible: http://physik.de/playouts.pdf What is the relationship between ELO as shown on the graph, and Go rating, in AGA or EGF terms? Nick I will now do some comparison with proportional increasing playouts in pachi and oakfoam. Detlef Am Montag, den 18.11.2013, 21:11 +0100 schrieb Petr Baudis: On Sun, Nov 17, 2013 at 03:11:22PM +0100, Erik van der Werf wrote: make sure Pachi isn't doing any kind of pondering in the background. Indeed, Pachi will ponder by default. Turn pondering off by passing pondering=0 on the commandline. pachi -d 0 -t =4000 -r chinese threads=1,max_tree_size=2048 Also, it may be worth passing pass_all_alive unless you are doing a sophisticated scoring procedure, to make sure Pachi captures all dead groups at the end of the game. P.S.: Do your results imply that on 4000 playouts/move, oakfoam is quite stronger than Pachi now? I'd love to hear more. :) How does the playout speed compare? ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Nick Wedd n...@maproom.co.uk ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Sylvain Gelly's thesis?
I found a copy there: http://bibliographie.jeudego.org/these_sylvain-gelly.pdf On 11/06/2013 10:51 AM, Ingo Althöfer wrote: Hello, is Sylvain Gelly's doctoral dissertation (from 2007) still available somewhere in electronic form? Thanks in advance, Ingo. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Monte-carlo simulations vs. MCMC
MCMC has little to do with what we do in computer Go. In MCTS we have a Markov Chain and we take Monte-Carlo samples from it, but the purpose is really not the same at all as what MCMC algorithms do. I recommend the wikipedia articles. It is difficult to really get an idea of MCMC by reading a general description. It is probably best that you get a feeling of what it is by studying the details of a real MCMC algorithm. The most basic MCMC algorithm is the Metropolis-Hastings algorithm: https://en.wikipedia.org/wiki/Metropolis%E2%80%93Hastings_algorithm Rémi On 1 nov. 2013, at 02:49, Darren Cook dar...@dcook.org wrote: I was reading this post How would you explain Markov Chain Monte Carlo (MCMC) to a layperson?: http://stats.stackexchange.com/q/165/5503 The first few answers confused me, definitely not layperson-ready I thought! But then these two talked about it in the context of board games, so I kind of got the idea: http://stats.stackexchange.com/a/438/5503 http://stats.stackexchange.com/a/12680/5503 But this just sounded like Monte-Carlo simulations. Which bit is the Markov-Chain? I thought I'd post here in the hope that someone could explain Markov-Chain Monte Carlo in computer go terms. Are we already using Markov-Chains in MCTS, just by another name? If not, why not? (I.e. is it an idea that was tried but didn't work very well for reasons we don't understand very well? Or there something about the nature of the go rules that mean it cannot be done? etc.) Thanks, Darren -- Darren Cook, Software Researcher/Developer http://dcook.org/work/ (About me and my work) http://dcook.org/blogs.html (My blogs and articles) ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Monte-carlo simulations vs. MCMC
On 1 nov. 2013, at 13:32, Darren Cook dar...@dcook.org wrote: MCMC has little to do with what we do in computer Go. In MCTS we have a Markov Chain and we take Monte-Carlo samples from it, but the purpose is really not the same at all as what MCMC algorithms do. I recommend the wikipedia articles. It is difficult to really get an idea of MCMC by reading a general description. It is probably best that you get a feeling of what it is by studying the details of a real MCMC algorithm. The most basic MCMC algorithm is the Metropolis-Hastings algorithm: https://en.wikipedia.org/wiki/Metropolis%E2%80%93Hastings_algorithm Thanks Remi. Quoting from that article: ...[a] method for obtaining a sequence of random samples from a probability distribution for which direct sampling is difficult. This sequence can be used to approximate the distribution (i.e., to generate a histogram) This sounds exactly like using N monte-carlo simulations at a node in an MCTS tree, generating a histograms of possible scores. The highest point on the histogram is used as the best-guess estimate of the score. When you have two peaks it implies some unstable situation, like a semeai. Etc. You mentioned the purpose is not the same. Can you elaborate? In MCMC the distribution is given to you with some kind of mathematical definition, and the challenge is to create a Markov Chain that approximates the distribution well. (If Markov-Chain is a nice clean synonym for rules of the game, whether the game is go or weather systems, I feel I am on home ground!) Darren ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Reminder: Handicap 29 prize
Hi Ingo, It seems mfgo1998 is not playing on KGS any more. Rémi On 6 oct. 2013, at 15:44, Ingo Althöfer wrote: Hello, my 1,000-Euro prize for a bot that beats the old Many Faces of Go (version without Monte Carlo components) at handicap 29 on the 19x19-board ist still open. When I announced the prize in January 2012, some people were optimistic to catch the money. But no challenger really came up, and from season to season my confidence grows that I will not have to pay the prize. Here are the details: http://www.althofer.de/handicap-29-prize.html Ingo. PS: As a human player, Martin Mueller from our scene had achieved the task in August 1998. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Computer Go Forum?
godiscussions has been broken for a long time. lifein19x19 is where people moved to. They have a computer-go section: http://www.lifein19x19.com/forum/viewforum.php?f=18 There is almost no programmer and no technical discussion there. On 2 oct. 2013, at 14:04, Ben Ellis wrote: I'd be surprised if there isn't a section on godiscussions. http://godiscussions.com/ it appears to be broken at the moment. On Sun, Sep 29, 2013 at 8:51 AM, Joshua Shriver jshri...@gmail.com wrote: Please excuse my various post, I'm coming from the chess world but with a long time interest in Go. Is there a computer go forum, kinda like talkchess.com but for this field of study? So far this forum has been my latest feed for information. -Josh ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] SGF format: How to specify prisoner count?
Yes, but I don't know the move history. What if I want to score a game from a photo of the final position? If I don't assume balanced passing, I must specify the number of prisoners (or number of extra passes). I think I can do it by adding pass moves to the sgf game, before the setup. That should indicate which player passed more. Rémi On 8 sept. 2013, at 10:18, Hideki Kato wrote: If B played 2 stone in a row, W must played a pass in between. Hideki Rémi Coulom: 31259b1d-7305-41b7-b223-fca7da22b...@free.fr: Hi, I wonder if it is possible to indicate the number of prisoners when setting up a position in SGF. I would need it for my image-capture app. The difference in the number of prisoners is important in Japanese rules. It is possible to guess its value by assuming that both players passed the same number of times (when handicap and color to move are known). But if one player passed more, then there is no way to guess it. I will probably ignore this issue and assume balanced passing when scoring japanese games. It should work well in most situations, but I don't like having to rely on this assumption. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Hideki Kato mailto:hideki_ka...@ybb.ne.jp ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Kifu-Snap: automatic go-board image recognition
Hi, A new version is available now. It supports automatic grid recognition (19x19 only). That was much more complicated to do than what I originally thought. Stone recognition was not improved, it is exactly the same as before. I am sorry I changed the package signature, so you'll have to uninstall the previous version in order to be able to install the new one. This new version will work until October 1st. I am sorry if you were hit by the limit of the previous version. I thought I would release an upgrade earlier. I thank all the users who sent photos. I don't really need more, but feel free to continue sending them to me. Have fun, and please let me know if you have any comment or suggestion. Rémi On 29 mai 2013, at 16:55, Rémi Coulom wrote: Hi, I have just put this on my web page, if some of you are interested: http://remi.coulom.free.fr/kifu-snap/ Kifu-Snap is an experimental tool to recognize a go board from a picture. On this web page you can download a free demo of the current Android prototype. You might find this prototype useful already, as it should let you score a finished game much faster than manual scoring. Still the recognition rate is far from perfect. The purpose of releasing this prototype is that the app will let you email your board pictures to me if you wish to do so. I need to collect a large database of varied board samples in order to train the machine-learning algorithms of kifu-snap better. So I'd appreciate your help. In exchange for this data, the next version of Kifu-Snap is likely to recognize your board better. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Kifu-Snap: automatic go-board image recognition
Hi, I have just put this on my web page, if some of you are interested: http://remi.coulom.free.fr/kifu-snap/ Kifu-Snap is an experimental tool to recognize a go board from a picture. On this web page you can download a free demo of the current Android prototype. You might find this prototype useful already, as it should let you score a finished game much faster than manual scoring. Still the recognition rate is far from perfect. The purpose of releasing this prototype is that the app will let you email your board pictures to me if you wish to do so. I need to collect a large database of varied board samples in order to train the machine-learning algorithms of kifu-snap better. So I'd appreciate your help. In exchange for this data, the next version of Kifu-Snap is likely to recognize your board better. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Kifu-Snap: automatic go-board image recognition
On 29 mai 2013, at 19:38, Gabriel Benmergui wrote: Nice work Remi. Would gladly put up a link on kaya for it. Im assuming it makes it into sgf? Yes, it produces an sgf file. Thanks for the link offer. The app. is currently very experimental, and maybe not ready for prime time, so I don't wish to advertise it too much. But feel free to link as you wish. A great usage for this would be automatic scribes. Yes. For those who are curious, I can give some references: Ananta Srisuphab An application for the game of Go: Automatic live Go recording and searchable Go database http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=6412186tag=1 Teemu Hirisimaki Extracting go game positions from photographs http://users.ics.aalto.fi/thirsima/gocam/gocam.pdf Alexander Seewald Automatic Extraction Of Go Game Positions From Images: A Multi-Strategical Approach to Constrained Multi-Object Recognition http://www.seewald.at/files/2007-04.pdf Steven Scher Making Real Games Virtual http://users.soe.ucsc.edu/~sscher/publications/ICPR2008-MakingRealGamesVirtual-Poster.pdf http://users.soe.ucsc.edu/~sscher/publications/ICPR2008-MakingRealGamesVirtual.pdf I found that this problem turns out to be considerably more difficult than what I thought it would be. Recognizing the board is so obvious for a human observer, it does not seem like a difficult task for a computer. But it is really difficult. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Can we kick out the conference spammers?
I tried to send a polite request to Rudy Lolo but he did not answer. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CLOP: Confident Local Optimization for Noisy Black-Box Parameter Tuning
That is difficult. Usually the central win rate is optimistic, and the win rate over all games is pessimistic. So those numbers should give you an idea of the range of the win rate you can expect with current estimated optimal parameters. But it might be possible to have narrower estimates. The problem is that this all depends on how quadratic the function is, which is difficult to determine. Rémi On 25 mars 2013, at 20:09, Petr Baudis wrote: Hi! On Sat, Jul 07, 2012 at 11:52:21PM -0700, Michael Williams wrote: Are the optimized values the Mean column on the Max tab? How does one get them out? Copy to clipboard only works for a single cell at a time. I'm on Windows. I'm wondering how to determine whether the parameters are already likely close to the optimal values. Can I use the winrate/Elo confidence bounds for that in some way? (Sorry if this has already been answered somewhere, I wasn't able to find it.) Petr Pasky Baudis ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CLOP: Confident Local Optimization for Noisy Black-Box Parameter Tuning
Hi, If you wish to explore the behaviour of CLOP on a particular problem, I recommend the test_swig.py script in the programs/clop/script/artificial directory. You have to edit this script to select the experiment you wish to run. First, select your problem. You can do this by uncommenting one (only) of the p = lines at the beginning of the script. For the Log function, uncomment p = CPLog1D() and comment out the other ones. Then, you can select the optimization method. The default is normal quadratic CLOP. You can also use cubic CLOP, CEM, or BAST if you wish. Finally, in the main program section you can select what you wish to do with this problem and method: - eb.threads will run many replications of the optimization experiment in parallel, and will produce some statistics - eb.gnuplot will run the optimization once and plot the samples. You can select the seed. - eb.multi_gnuplot will plot the optimization for many seeds in succession - eb.tikz: probably produces something similar to eb.gnuplot in tikz format for incorporation into LaTeX documents Rémi On 21 mars 2013, at 09:19, Chin-Chang Yang wrote: Thank you, Remi and Olivier. I implemented the test functions in my project, and tested the performance of my optimizers, which reached the regret to 1e-3 with 1e7 samples in the 1-dimensional LOG function. In the CLOP paper, it reports that CLOP can reach the regret to 1e-5 with 1e7 samples. It is very interesting that how CLOP can perform so well. I am going to trace the CLOP codes by gdb, but I cannot find the script that can reproduce the experimental results in Fig.4 (a). I guess the script is located in programs/clop/script/artificial but I am not sure which script can reproduce the experiments. I will be glad if anyone can give me some hints. Best regards, Chin-Chang Yang, 2013/3/21 -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go-boun...@dvandva.org] On Behalf Of Remi Coulom Sent: Wednesday, March 6, 2013 5:37 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] CLOP: Confident Local Optimization for Noisy Black-Box Parameter Tuning Yes. f(x) is not the output. The output is either 0 or 1, and f(x) is the probability of 1. Rémi On 6 mars 2013, at 09:04, Chin-Chang Yang wrote: Thank you, Olivier. Let the observable function value be o(x). It can be defined as: o(x) = 1, with probability f(x); o(x) = 0, with probability (1 - f(x)). where f(x) = 1 / (1 + e(-r(x))) has been defined in the paper. Also, we can see that the expected value is f(x). Did I get this correct? Best regards, Chin-Chang Yang, 2013/03/06 2013/3/6 Olivier Teytaud teyt...@lri.fr It's a Bernoulli noise. define f (x) = 1/ (1 + e(−r(x)) ) and the objective function at x is 1 with probability f(x). So the expected value at x is f(x), but the values you get are noisy. Best regards, Olivier Since the functions are not noise-free, they should be defined in terms of some noise. I really need the definition of the noise for comparison between CLOP and other optimizers. I have downloaded the source codes, but I cannot find the codes related to the noise currently. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Chin-Chang Yang ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] parameter optimization (clop others)
If I understand your notation correctly, CLOP can get C = 1/3 for smooth functions. In theory, regression with a polynomial of degree d gets C = 0.5 * d / (d + 1) if the function is as smooth as the polynomial with bounded (d+1)th derivative. There is an experiment in the CLOP paper that confirms emprically the C = 3/8 for d = 3. But these are asymptotic rates that may not be very relevant in practice. The CLOP paper has references to papers with these bounds (Chen-1988 for the proof that it is not possible to do better. IIRC, some variations of stochastic gradient have been proved to match the bound, so the bound is tight). Rémi On 7 févr. 2013, at 13:00, Olivier Teytaud wrote: Hi; as many of you I guess, I often try to optimize parameters of strategies. In games, if I optimize against a fixed opponent, this is a noisy optimization problem: I want the parameters with the best success rate against the chosen opponent, which can be evaluated thanks to repeated (time-consuming) games. The state of the art in noisy optimization is slightly unreadable. For most tools there are properties such as log( distance to optimum ) ~ - C log( number of evaluations) for some positive C. In some cases, C=1/2 (or something close to 1/2 depending on derivability conditions), but this is restricted to easy problems. For other families of functions and under some technical assumptions aimed at getting rid of too simple objective functions, Basically, one then get rates such as C=1/4, for a quadratic objective function. However, the bounds are C=1/2, so there is still a gap. So basically one can get nearly any result depending on the assumptions :-) I'd like to know which rates you get on your favorite optimization problems. Maybe many people here don't care about noisy optimization from a maths point of view, but I'm pretty sure that people here work on real noisy optimization problem and if they plot curves such as the equation above it will provide interesting information. Thanks for any feedback :-) Best regards, Olivier ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] parameter optimization (clop others)
On 7 févr. 2013, at 21:01, Olivier Teytaud wrote: (the constant term in the linear slope in log-log term depends on the number of parameters, this might be relevant for Brian's remark on the number of parameters). I am not sure it does. In fact I would guess it does not. If the optimization time is long enough, tuning many parameters simultaneously is probably faster than adding them one by one. The big problem is if the optimization time is long enough. If you wish to optimize parameters with a finite budget (overnight or over a week), then it is better to tune very well a few of them rather than tune badly many of them. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Board Sizes
On 01/07/2013 03:02 PM, Vlad Dumitrescu wrote: Hi! On Mon, Jan 7, 2013 at 2:51 PM, Don Dailey dailey@gmail.com mailto:dailey@gmail.com wrote: 2013/1/7 Don Dailey dailey@gmail.com mailto:dailey@gmail.com I have a question concerning a related question, mirror go.On an even board the second player can presumably just mirror the opponent and never lose but I am not sure I buy that. I am a very weak player myself but is it true that cannot force your opponent into a bad move if he mirrors you? And if mirror go is a valid way to stay even couldn't a pro just sacrifice something in the center to break the symmetry? I'm talking about even boards, not odd boards. I know that odd boards cannot be mirrored indefinitely. It's easy to come up with a way to force breaking the symmetry: play the center as a cross-cut and black then gives atari: if white gives atari too, black captures and white cannot do likewise because of the missing captured stone. So white is forced to protect the atari and symmetry is broken. Black can then get a better position in the center in sente. regards, Vlad But you can't get a cross-cut in the center with mirror go. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] 2013 Computer Olympiad in Yokohama, Japan
http://www.grappa.univ-lille3.fr/icga/event.php?id=44 Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Go software on Mac OS X
You can use gogui. On 23 nov. 2012, at 14:28, Aja Huang wrote: Dear all, Is there any Go software on OS X like MultiGo on Windows that I can view and edit .sgf game records? Thanks, Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Develop a client for KGS?
On 8 nov. 2012, at 14:09, Don Dailey wrote: I thought KGS had their own proprietary protocol? Yes. And I don't think GTP can be of any help to connect a web client to KGS. See that page: http://senseis.xmp.net/?KGSueMe%2FDiscussion Note that wms himself is developing a web client: https://plus.google.com/108736506961432085848/posts/8VJ6qrq7stG Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] TAAI 2012 Computer Go Tournaments
On 2 oct. 2012, at 09:18, Jirong wrote: Dear all, We are happy to announce TAAI 2012 Computer Go Tournaments. The top bot can play with pro go players. If you are interested, please register at http://www.tcga.tw/taai2012/eng/ - TAAI 2012 Computer Go 19x19 Tournament 1. Board size: 19x19. 2. Time system: 15 minutes for each side, Byo-yomi 10 moves in 30 seconds. 3. Komi: 7.5 points. 4. Tournament system: Quadruple Round-robin. 5. Score: The winning bot gets 1 points and each bot gets 0.5 point if the result is draw. Can the result be a draw? Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] TAAI 2012 Computer Go Tournaments
On 2 oct. 2012, at 11:30, Aja Huang wrote: 5. Score: The winning bot gets 1 points and each bot gets 0.5 point if the result is draw. Can the result be a draw? Rémi On September 5, two top Go players Lee Sedol and Gu Li played a game ended in a quadruple ko, see http://gogameguru.com/quadruple-ko-group-of-death-17th-samsung-cup/ Perhaps Jirong was influenced by that interesting event. :) I believe the KGS superko rules would not consider it as a draw. It is just like another superko. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] An unusual seki
A link to the file might be more convenient for many: http://files.gokgs.com/games/2012/8/14/Blubbel-AyaBot4-2.sgf Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CLOP: Confident Local Optimization for Noisy Black-Box Parameter Tuning
If you can edit the source code and re-compile, you can try replacing: Qt::ItemIsEnabled by Qt::ItemIsEnabled | Qt::ItemIsSelectable in MainWindow.cpp I don't have time to test or prepare a new version, sorry. Rémi On 8 juil. 2012, at 08:52, Michael Williams wrote: Are the optimized values the Mean column on the Max tab? How does one get them out? Copy to clipboard only works for a single cell at a time. I'm on Windows. On Thu, Sep 1, 2011 at 3:01 AM, Rémi Coulom remi.cou...@free.fr wrote: Hi, This is a draft of the paper I will submit to ACG13. Title: CLOP: Confident Local Optimization for Noisy Black-Box Parameter Tuning Abstract: Artificial intelligence in games often leads to the problem of parameter tuning. Some heuristics may have coefficients, and they should be tuned to maximize the win rate of the program. A possible approach consists in building local quadratic models of the win rate as a function of program parameters. Many local regression algorithms have already been proposed for this task, but they are usually not robust enough to deal automatically and efficiently with very noisy outputs and non-negative Hessians. The CLOP principle, which stands for Confident Local OPtimization, is a new approach to local regression that overcomes all these problems in a simple and efficient way. It consists in discarding samples whose estimated value is confidently inferior to the mean of all samples. Experiments demonstrate that, when the function to be optimized is smooth, this method outperforms all other tested algorithms. pdf and source code: http://remi.coulom.free.fr/CLOP/ Comments, questions, and suggestions for improvement are welcome. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CLOP: Confident Local Optimization for Noisy Black-Box Parameter Tuning
After all, I found a little time to try it. Unfortunately it does not work. I would have to implement my own self-made copy function like explained on that web page: http://www.qtcentre.org/threads/11090-Copy-row%28s%29-from-QTableWidget I added it to the TODO list. Rémi On 8 juil. 2012, at 12:34, Rémi Coulom wrote: If you can edit the source code and re-compile, you can try replacing: Qt::ItemIsEnabled by Qt::ItemIsEnabled | Qt::ItemIsSelectable in MainWindow.cpp I don't have time to test or prepare a new version, sorry. Rémi On 8 juil. 2012, at 08:52, Michael Williams wrote: Are the optimized values the Mean column on the Max tab? How does one get them out? Copy to clipboard only works for a single cell at a time. I'm on Windows. On Thu, Sep 1, 2011 at 3:01 AM, Rémi Coulom remi.cou...@free.fr wrote: Hi, This is a draft of the paper I will submit to ACG13. Title: CLOP: Confident Local Optimization for Noisy Black-Box Parameter Tuning Abstract: Artificial intelligence in games often leads to the problem of parameter tuning. Some heuristics may have coefficients, and they should be tuned to maximize the win rate of the program. A possible approach consists in building local quadratic models of the win rate as a function of program parameters. Many local regression algorithms have already been proposed for this task, but they are usually not robust enough to deal automatically and efficiently with very noisy outputs and non-negative Hessians. The CLOP principle, which stands for Confident Local OPtimization, is a new approach to local regression that overcomes all these problems in a simple and efficient way. It consists in discarding samples whose estimated value is confidently inferior to the mean of all samples. Experiments demonstrate that, when the function to be optimized is smooth, this method outperforms all other tested algorithms. pdf and source code: http://remi.coulom.free.fr/CLOP/ Comments, questions, and suggestions for improvement are welcome. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Nice and Hyperthreading
Do you mean you have N real cores (so 2N hyperthreads), and run your program with N threads? You might be able to disable hyperthreading in the BIOS to solve this problem. I imagine that if you use 2N threads in your program, it should almost stop the clop experiment. Rémi On 29 mai 2012, at 09:39, ds wrote: Hi, a little bit of topic, but I hope some of you had a simelar problem: Before the age of Hyperthreading (this was two weeks ago for me) I could run clop with a high nice value on my computer and run the kgs go program with usual value. This way I could use the computational power which is not needed to play to do my optimization of parameters. Now with Hyperthreading even niced processes harm the performance a lot. I understand, that this is a conceptional problem, as the operating system does not know that using virtual core 0 harms virtual core 4, but this knowledge does not help me:( Is this kind of usage really so unusual now, that there is no solution? Thanks a lot Detlef ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] New commercial Crazy Stone
On 21 mai 2012, at 08:17, Chin-Chang Yang wrote: Hi, I want to know more information about English version of CrazyStone 2012. Can anyone answer the following questions? Does it support Chinese rule? Yes. What does the Histogram mean? It is the distribution of playout scores. In Record Analysis List window, what does the Situation and Delta mean? Situation is the expected score. Delta is the difference in win rate between the move suggested by Crazy Stone, and the move actually played during the game. Can I set the amount of playouts in Analysis Mode to maximize its accuracy? You can set the thinking time for analyzing the whole game. You can set it from 5 minutes to 8 hours, if I remember correctly. And, can I set the amount of playouts in Game Mode to control computer's strength? You can choose a level. The web site says 10 levels of play. The most recent beta I tested actually had 20, so it may be 20 after all. Each level corresponds to a total number of playouts per game. Thank you. Best regards, Chin-Chang Yang, 2012/05/21 2012/5/17 Ingo Althöfer 3-hirn-ver...@gmx.de Hello Remi, Because there was so much demand for analysis features, Unbalance reconsidered their plans, and after all made an English version with analysis features. It is even better than the Japanese version now, because it also shows the histogram. You can see screenshots there: http://www.unbalance.co.jp/igo/eng/ Rémi Thanks for the information. One question. On the website it is said To play Crazy Stone 2012, your PC must be connected to the Internet. How is this sentence meant? Do I need internet connection for installing the program or do I need internet connection every time when I want to play? Regards, Ingo. -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Chin-Chang Yang ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] New commercial Crazy Stone
On 21 mai 2012, at 09:36, Rémi Coulom wrote: You can choose a level. The web site says 10 levels of play. The most recent beta I tested actually had 20, so it may be 20 after all. Each level corresponds to a total number of playouts per game. There are in fact 10 levels as indicated in the web site, sorry for the confusion. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] A Regression test set for exploring some limitations of current MCTS programs in Go
Now with the correct e-mail address. On 17 mai 2012, at 16:43, Rémi Coulom wrote: I took a closer look at the games. 19 is hanezeki: http://senseis.xmp.net/?Hanezeki I don't worry too much about that. Did this ever occur in a real game? I would recommend using non-integer komi for your tests, because they test the ability of the program to deal with jigo at the same time as they test seki. Dealing with jigo in the search is not an easy job: it is much more difficult to get a consistent search, with proved convergence to optimal play, when the outcome of the game is not binary. Completely greedy search will solve any position with non-integer komi, but it is likely to fail with integer komi (ie, get stuck on jigo when a stronger move can win but has a low evaluation in the beginning of the search). Crazy Stone evaluates hanezeki correctly if komi is set to 7.5 instead of 7.0. Sorry, that should be 6.5. With 6.5, Crazy Stone still fails. So hanezeki is still difficult. Rémi case11 is strange. In the variation contained in the sgf, W loses by two points. Aja, are you sure case11 is correct? Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] detecting cross inside an eye
I suppose he means an empty intersection. The rule used usually is: all 4 immediate neighbours are of my color and the opponent has no more than two (one on the side) stone(s) in diagonal. That rule will improve the strength of stop-cur considerably. Rémi On 10 mai 2012, at 18:55, Álvaro Begué wrote: What do you mean by cross? On Thu, May 10, 2012 at 12:11 PM, folkert folk...@vanheusden.com wrote: Hi, How do you detect that a cross is inside an eye? (eye? - circle, connected chain of pieces is what i mean) Folkert van Heusden -- Multitail es una herramienta flexible que permite visualizar los log file y seguir la ejecución de comandos. Permite filtrar, añadir colores, combinar archivos, la visualización de diferencias (diff- view), etc. http://www.vanheusden.com/multitail/ -- Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] cgos client crash: too many open files
After a few days of running my program on CGOS, the 9x9 client crashed with the
error message below. lsof shows that tclsh has a lot of open files. Maybe
there is something that needs to be closed somewhere. Anybody got the same
problem? Maybe that only happens when there are many players and the client
re-opens a new file each time there is a new player.
Rémi
Server: cgos.boardspace.net
Port: 6867
PLAYER PRIORITY
--
CS-1250k 25.0%
CS-250k25.0%
CS-50k 25.0%
CS-10k 25.0%
New player: CS-1250k
couldn't execute ./cs9-1250k.sh: too many open files
while executing
open |$invoke r+
(procedure openPlayer line 7)
invoked from within
openPlayer $name $password $invoke
(procedure switchPlayer line 49)
invoked from within
switchPlayer
(gameover arm line 14)
invoked from within
switch -exact $cmd {
info {
# informational mssage from CGOS only
# ---
log [string range $s 5...
(while body line 27)
invoked from within
while {1} {
set err [catch {set sock [socket $cgos_server $cgos_port]} msg]
if { $err } {
log Server startup return code: $err msg: $msg...
(file ./cgosGtp.tcl line 402)
output of lsof:
(400 similar lines snipped)
tclsh 31785coulom 400u sock 0,49589142 can't
identify protocol
tclsh 31785coulom 401u sock 0,49595353 can't
identify protocol
tclsh 31785coulom 402u sock 0,49596062 can't
identify protocol
tclsh 31785coulom 403u sock 0,49600736 can't
identify protocol
tclsh 31785coulom 404u sock 0,49603055 can't
identify protocol
tclsh 31785coulom 405u sock 0,49608415 can't
identify protocol
tclsh 31785coulom 406u sock 0,49610568 can't
identify protocol
tclsh 31785coulom 407u sock 0,49611420 can't
identify protocol
tclsh 31785coulom 408u sock 0,49613458 can't
identify protocol
tclsh 31785coulom 409u sock 0,49617849 can't
identify protocol
tclsh 31785coulom 410u sock 0,49623717 can't
identify protocol
tclsh 31785coulom 411u sock 0,49625689 can't
identify protocol
tclsh 31785coulom 412u sock 0,49628467 can't
identify protocol
tclsh 31785coulom 413u IPv4 9628958TCP
cluster-mostrare.lille.inria.fr:35136-ip14.50-31-1.static.steadfastdns.net:6819
(CLOSE_WAIT)
tclsh 31785coulom 414u IPv4 9631218TCP
cluster-mostrare.lille.inria.fr:56096-ip14.50-31-1.static.steadfastdns.net:6819
(CLOSE_WAIT)
tclsh 31785coulom 415u IPv4 9634615TCP
cluster-mostrare.lille.inria.fr:50507-ip14.50-31-1.static.steadfastdns.net:6819
(ESTABLISHED)
tclsh 31785coulom 416w FIFO 0,69634536 pipe
tclsh 31785coulom 417r FIFO 0,69634537 pipe
tclsh 31785coulom 419u REG 8,8 0 34137
/tmp/tcleTg8Cq (deleted)
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] cgos
I reserved a small 4-core node of our cluster to connect 9 fast instances of Crazy Stone, 3 for each board size. On 23 avr. 2012, at 19:48, ds wrote: Hi, would be nice, if one could arrange a typical date, where one can hope to find more programs on cgos. E.g. once a week or once every two weeks on monday? Depending on the opponents one could let the programs run for some hours or some days. For the last weeks it was quite empty with only the standard programs running. You can only get rated around 1800ELO with some accuracy... Detlef ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] cgos
In Crazy Stone, I set the total number of playouts per game, not per move. 250k per game is similar in strength to 2k playouts per move (a bit weaker, maybe). But it takes less time overall for one game (because I apply time-management heuristics to playouts), which allows faster testing. Rémi On 24 avr. 2012, at 18:38, ds wrote: Sorry, I do not want to get all your secrets, but how do you do 250k playouts with seconds for a whole game on all boardsize with 4 cores? Or does reservation mean a minimum reservation? Detlef Am Dienstag, den 24.04.2012, 12:40 +0200 schrieb Rémi Coulom: I reserved a small 4-core node of our cluster to connect 9 fast instances of Crazy Stone, 3 for each board size. On 23 avr. 2012, at 19:48, ds wrote: Hi, would be nice, if one could arrange a typical date, where one can hope to find more programs on cgos. E.g. once a week or once every two weeks on monday? Depending on the opponents one could let the programs run for some hours or some days. For the last weeks it was quite empty with only the standard programs running. You can only get rated around 1800ELO with some accuracy... Detlef ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] cgos
In fact I have just figured out that there are problems with counting playounts when the number is low. I did not notice it in the past because I usually test with at least 250k. But I also have a minimum number of playouts per move that it will always search. The 10k versions hit this minimum, and may in fact search more than 10k playouts in a game. Rémi On 24 avr. 2012, at 20:33, Aja Huang wrote: On 9x9 Erica now scores about 85% on 300 playouts / move against GnuGo 3.8 Level 10, around of rating 2100. From current CGOS results looks like Crazy Stone is much stronger. :) Aja 2012/4/24 ds d...@physik.de Thanks a lot, there is a long way to go for us with this impressible strength:) Detlef Am Dienstag, den 24.04.2012, 19:43 +0200 schrieb Rémi Coulom: In Crazy Stone, I set the total number of playouts per game, not per move. 250k per game is similar in strength to 2k playouts per move (a bit weaker, maybe). But it takes less time overall for one game (because I apply time-management heuristics to playouts), which allows faster testing. Rémi On 24 avr. 2012, at 18:38, ds wrote: Sorry, I do not want to get all your secrets, but how do you do 250k playouts with seconds for a whole game on all boardsize with 4 cores? Or does reservation mean a minimum reservation? Detlef Am Dienstag, den 24.04.2012, 12:40 +0200 schrieb Rémi Coulom: I reserved a small 4-core node of our cluster to connect 9 fast instances of Crazy Stone, 3 for each board size. On 23 avr. 2012, at 19:48, ds wrote: Hi, would be nice, if one could arrange a typical date, where one can hope to find more programs on cgos. E.g. once a week or once every two weeks on monday? Depending on the opponents one could let the programs run for some hours or some days. For the last weeks it was quite empty with only the standard programs running. You can only get rated around 1800ELO with some accuracy... Detlef ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] citeulike reminder
Hi, I have just noticed that the citeulike collection of bibliographic references has not been updated in a while: http://www.citeulike.org/group/5884/library I'd like to invite authors of papers related to computer go to post them there. That will improve their visibility. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] New commercial Crazy Stone
On 8 avr. 2012, at 22:13, Ingo Althöfer wrote: This product uses the newest version and, in addition, has analysis window feature, which is very useful for anaylizing and research of games. We need it in Western version, too. I am sorry this won't be in the western version. You can simply play games against the program, not analyze game records. But the new version plays in a beautiful 3D traditional Japanese room with tatami. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] New commercial Crazy Stone
Hi Ingo, I don't know the exact date, but the English version will be available soon. The old version is available at a discount until then. The new version will be available from that page: http://www.unbalance.co.jp/igo/eng/ It currently says in the middle of April. Rémi On 8 avr. 2012, at 20:01, Ingo Althöfer wrote: Hello, in a german go forum someone gave the following link http://www.unbalance.co.jp/igo/sigo2012/ and wrote that this is about a new commercial version of Crazy Stone. Unfortunately I can not read the Japanese letters. Can someone here translate the text or give the information in his own words? Will there also be a new western version of commercial Crazy Stone? Cheers, Ingo. -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to pachi2!
Thanks Nick. Crazy Stone's analysis of the victory of Zen against Pachi is still online: http://www.grappa.univ-lille3.fr/~coulom/CrazyStone/2012-03-14-Pachi/index.html Rémi On 22 mars 2012, at 12:41, Nick Wedd wrote: Congratulations to pachi2, winner of the Slow KGS bot tournament! My report is a week late, because I have been away on holiday (a trip to Scotland, taking in the Isle of Skye Go tournament). It is finally available at http://www.weddslist.com/kgs/past/S12.1/index.html I hope you will email me about mistakes in it as usual. The report explains how MyGoFriend lost a game by Forfeit, when its operator logged in using its account to fix a problem. This was not my decision, it is a feature of the server's tournament-handling code. So please, if your bot misbehaves, be aware that logging in with its account will result in a forfeited game. Nick -- Nick Wedd n...@maproom.co.uk ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] useless ko threats
Accelerated UCT does this: https://www.conftool.net/acg13/index.php/Hashimoto-Accelerated_UCT_and_Its_Application_to_Two-Player_Games-111.pdf?page=downloadPaperfilename=Hashimoto-Accelerated_UCT_and_Its_Application_to_Two-Player_Games-111.pdfform_id=111form_version=final Rémi On 6 mars 2012, at 20:50, Stefan Kaitschick wrote: On Tue, Mar 6, 2012 at 3:16 PM, Álvaro Begué alvaro.be...@gmail.com wrote: I think the solution is the introduction of some flavor of minimax into the tree search. For instance, once a node has been visited more than a certain number of times, the score that we'll back out from it is just the score of the best child, instead of an average that might be polluted by bad moves. Of course someone else would have to figure out the details of how to do this. :) Álvaro. This isn't a go specific problem, is it. But that means that either this is a wild goose chase, or there is room for improvement of general MCTS here. Winrate is certainly the legitimate criterion to distinguish between siblings. But should the parent node really get full credit/blame for the weaker continuation (in the case of go refutation) attempts? This is really one side of an extreme, minimax being the other side. Maybe something in between could be useful. The weight of a node could grow with greater than linear speed in respect to the number of attempts made. Stefan ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] About Minirock
Hi Xaryl, You don't have to apologize. You caused no trouble. Welcome to the list. Rémi On 14 févr. 2012, at 17:11, Xaryl C wrote: First of all, sorry for creating a new thread. I wasn't on the mailing list, and wasn't sure how to respond to a thread without the original mail. Also sorry that I wasn't aware of the discussion until now. I am the one responsible for the bot. I started coding a go bot when school was boring, that was when classical bots prevailed. Then life got busy, I tried MC, made some nice progress in terms of strength, but never even got the chance to properly debug the multi threading codes. Then sometime ago, Minirock (the player) wanted to have a bot, so I hackishly glued (with macros and mass code replacing with scripts) and replaced the playout codes and search distribution along with tactical reading codes and pattern matchers onto Pachi's codes. Essentially, it is a bot with borrowed codes bloated with codes for classical bots organized in a way that would give computer science teachers heart attacks. The future is uncertain, I might or might not spend much time on it after resolving the serious bugs. It was mostly just for fun, and I didn't realize a weak random bot would caught attention. But hey, hi everyone, nice to meet you all and sorry for the trouble I caused. Xaryl p.s. I don't speak much German, but I will admit that I find the name Minirock a bit funny, so I guess that would be the bot's name from now on. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] cgos game archives
Hi, I'd like to get a copy of the game records of Crazy Stone on CGOS, and I cannot find a convenient way to do it. The archives do not seem up to date. Any suggestion? I'd be happy if I could at least get the archive of the January games. Thanks, Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] minirock2
I tried 3 games with Crazy Stone: http://www.gokgs.com/gameArchives.jsp?user=bonobot Each time reducing CPU. First game was full power, second game was 4 cores, third game was 2 cores and very fast play. 3rd game was close. Minirock2 looks rather strong. Info now says author is minirock. Maybe we could ask questions to minirock. Rémi On 28 janv. 2012, at 22:30, Hiroshi Yamashita wrote: Sorry, my guess is wrong. I tested 3 games vs Minirock2. Aya won 2 games and lost one. http://www.gokgs.com/gameArchives.jsp?user=ayamc4 But Minirock2 runs on 2 cores 2.8GHz, and Aya is on 6 cores 3.3GHz. I think Minirock2 is 1d or 2d, and if it uses faster machine, 2d or 3d. Zen19 also played 2 games vs Minirock2, and Zen won 2 games. http://www.gokgs.com/gameArchives.jsp?user=zen19 Hiroshi Yamashita - Original Message - From: Erik van der Werf erikvanderw...@gmail.com To: computer-go@dvandva.org Sent: Thursday, January 19, 2012 7:42 AM Subject: Re: [Computer-go] minirock2 No, not Steenvreter. Erik On Mon, Jan 16, 2012 at 12:46 PM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: I thought it is Steenvreter. Hiroshi Yamashita - Original Message - From: Ingo Althöfer 3-hirn-ver...@gmx.de To: computer-go@dvandva.org Sent: Monday, January 16, 2012 6:12 PM Subject: Re: [Computer-go] minirock2 Interesting speculation. Independent of the programmer question I woukd like to see Minirock2 playing rated games. Ingo. Original-Nachricht Datum: Mon, 16 Jan 2012 09:33:32 +0100 Von: Rémi Coulom remi.cou...@free.fr An: computer-go@dvandva.org Betreff: [Computer-go] minirock2 Anybody knows what is minirock2? Was it programmed by rock2? That would be big news :-) Fabien, are you reading? Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Game 2 goes to Zen: 1-1
In the most recents Crazy Analysis, Delta is the difference between the evaluation of this move, and the evaluation of the next position of the game. For instance, in this page: http://www.grappa.univ-lille3.fr/~coulom/CrazyStone/ingo/index.html At move 72, delta = 0.0431076 is 0.441597 (value of K12) - 0.39849 (value of next position) It is more accurate, because the next position is always searched with more playouts than the alternative move when analyzing the previous position. Rémi On 26 janv. 2012, at 14:16, Ingo Althöfer wrote: Hi Remi, I have one more question on CrazyAnalysis. What is the exact meaning of delta? In the detailed analysis on the corresponding move, are there two values where delta is the exact difference of them? I did not find any. Regards, ingo. -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] KGS, gogui and handicap stones
The problem is that kgs saves handicap stones as black moves (B), instead of setup stones (AB), when playing with Chinese rules. The sgf specification says that they should be setup stones. Gogui correctly sends a set_free_handicap command to your engine if setup stones are indicated instead of black moves. A solution is to edit the sgf files of KGS to properly use the AB property instead of B moves for handicap stones. Maybe gogui could be modified to consider the first B moves as handicap stones when the HA property is set with Chinese rules, and there is no setup stones. The best would be that wms fixes the sgf of KGS, of course. Rémi On 21 janv. 2012, at 11:25, ds wrote: Hi, sounds simple, but I could not figure out: If I load a game from kgs with gogui, my engine is not informed about the handicap stones. Therefore it counts wrong? Is there a workaround? Thanks Detlef ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Computer Go and EGC 2012
On 19 janv. 2012, at 20:49, Richard J. Lorentz wrote: After reading all these posts I have concerns about this whole game within a game idea. It's beginning to look like the pro's behavior becomes part of the experiment, which you probably don't want as it would be demeaning to the pro. I suspect the original spirit was to make it as attractive as possible to the pro so that he will be likely to participate. This might be best done by offering the most money you can with what you think is the fairest handicap. Thanks Richard for your message. I agree 100%. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Game 2 goes to Zen: 1-1
It is now replaced by the analysis of game 4: http://www.grappa.univ-lille3.fr/~coulom/CrazyStone/index.html I improved my system a lot in one day. Smoothness of the histogram depends on the number of playouts, not safety. This one is very smooth because it was running on a big machine instead of my laptop. Congrats to Zen, btw. Rémi On 16 janv. 2012, at 11:34, Isaac Deutsch wrote: For people who missed it in the Kibitz, Rémi Coulom also put together a very nice automatic analysis tool for this game: http://www.grappa.univ-lille3.fr/~coulom/CrazyStone/CrazyAnalysis.html As he stated, it's very experimental (it seems the Next link is sometimes broken??). I still find it very interesting to look at, especially the score histogram. I like how you can observe the gentle drift of the median to the right as the game progresses. It's interesting that sometimes the distribution is very smooth, while other times, it's rough and jagged. I guess this is related to the safety of groups and/or territory. Rémi, will you have further analyses for the upcoming games? I'm sure it would be greatly appreciated. Thank you! Cheers, Isaac Am 16.01.2012 um 08:07 schrieb Ingo Althöfer: Hello Jeff, thanks for setting up the relay, and also for posting the sgf with the comments here. In the other relay (tromp vs zen19n) John made a few comments directly after the game. I quote him here: N7 assumed black would respond to P4 at Q4. felt already lost when playing N7 [at move 48] I guess L7 [at move 46; IA] was already a little desperate. I have uploaded automatic analysis data by ManyFaces. * MC percents at http://www.althofer.de/tromp-zen-03-percents.jpg * Expected Territories at http://www.althofer.de/tromp-zen-03-territory.jpg Ingo. Original-Nachricht Datum: Sun, 15 Jan 2012 21:25:47 -0500 Von: Jeff Nowakowski j...@dilacero.org An: computer...@computer-go.org Betreff: Re: [Computer-go] Game 2 goes to Zen: 1-1 Game 3, with mostly dan-level comments: http://files.gokgs.com/games/2012/1/16/PaperTiger-2.sgf Zen won by resignation. Series is 2-1 in Zen's favor. I got Aja 6d and gogonuts 5d to kibitz, and participation by observer count was much better, around 60 versus the 260 or so for the unmoderated game. On 01/15/2012 03:03 PM, Jeff Nowakowski wrote: On 01/15/2012 10:18 AM, Michael Williams wrote: Kibitzing anything in that room is pointless for all the noise. For both games, I set up a clone game with moderated chat where only dans could kibitz. Unfortunately, at best it only had 10% participation (based on observer counts) and the strongest and most frequent kibitzers, Aja and gogonuts, kept to the unmoderated game. The one exception was Avidya (David Ormerod from gogameguru.com), who provided excellent comments. I'll try and recruit the high level dans before tonight's game, and see if I can get the moderated game listed on the Active Games tab next to the unmoderated one. If it doesn't work out, I'll stop trying to host these, as it takes some effort. It would be nice if this could be an automated and standard feature of KGS events, but I don't see that happening any time soon. Here are the two dan-kibitz games: http://files.gokgs.com/games/2012/1/14/PaperTiger.sgf http://files.gokgs.com/games/2012/1/15/PaperTiger.sgf They are a nice way to review the games with some high-level comments. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- NEU: FreePhone - 0ct/min Handyspartarif mit Geld-zurück-Garantie! Jetzt informieren: http://www.gmx.net/de/go/freephone ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Game 2 goes to Zen: 1-1
On 17 janv. 2012, at 11:05, Ingo Althöfer wrote: Hello Remi, thanks a lot for your nice service. Hopefully, this feature will flow also in forthcoming commercial versions of CrazyStone. Hi Ingo, Since many people seem so interested, we may try to include it in the next version. A few proposals for improvement: * You should mention that in each row the move itself can be clicked to get more information. * P(Black to win) might be given in percent, with only one digits behind the dot. For instance 48.2 instead of 0.482298 for position 0. Deleting the other digits improves readability. * Also for the delta-coulom I would prefer % with TWO digit behind the dot, so for instance 2.68 instead of 0.0267718. * For easier live readability (during the game) reverse order of the moves would be nice, having the most recent moves in the top lines. Thanks for your suggestions (and thanks to others that made comments, too). Normally, I should not worry too much about that, and let Unbalance polish the final look. There is a beginner game analyzed with the same tool: http://www.grappa.univ-lille3.fr/~coulom/CrazyStone/perceval/index.html I added a list ordered by delta, and player to move as background color. Thanks for thinking about it. By the way, I like the colours for the territories and also the score histogram. Maybe, one day such score histograms will help in the design of tools like dynamic komi. That's an interesting idea. I tried many ways to adjust dynamic komi, and I did use the histogram to choose it. Unfortunately, I never found any scheme that works. Changing komi during a search while keeping the same search tree always has catastrophic consequences. And throwing the tree away when komi is changed is a waste. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] minirock2
Anybody knows what is minirock2? Was it programmed by rock2? That would be big news :-) Fabien, are you reading? Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] kgs bot interface question
I have this problem too. I think the problem is not resignation. The problem is when several player challenge at the same time. You can tell kgsgtp to write a log. Below is such a log for CrazyStone. So it must be a bug in kgsgtp. Maybe wms reads this. Otherwise, we could send a message to him. Rémi 30 déc. 2011 01:32:12 com.gokgs.client.gtp.GtpClient c FIN: Game ended. Starting another. 30 déc. 2011 01:32:12 com.gokgs.client.gtp.GtpClient a PLUS FIN: No games to join. Creating an open game. 30 déc. 2011 01:32:12 com.gokgs.client.gtp.f a PLUS FIN: Joined challenge 30 déc. 2011 01:32:20 com.gokgs.client.gtp.f a PLUS FIN: Got challenge from xiaosugi, testing engine response. 30 déc. 2011 01:32:20 com.gokgs.client.gtp.GtpClient d LE PLUS FIN: Command sent to engine: boardsize 19 30 déc. 2011 01:32:20 com.gokgs.client.gtp.GtpClient d LE PLUS FIN: Got successful response to command boardsize 19: = 30 déc. 2011 01:32:20 com.gokgs.client.gtp.f a LE PLUS FIN: Board size 19 is acceptable 30 déc. 2011 01:32:20 com.gokgs.client.gtp.GtpClient d LE PLUS FIN: Command sent to engine: time_settings 135 0 0 30 déc. 2011 01:32:20 com.gokgs.client.gtp.GtpClient d LE PLUS FIN: Got successful response to command time_settings 135 0 0: = 30 déc. 2011 01:32:20 com.gokgs.client.gtp.f a LE PLUS FIN: Time system is acceptable 30 déc. 2011 01:32:20 com.gokgs.client.gtp.f a LE PLUS FIN: Sending challenge back to server 30 déc. 2011 01:32:20 com.gokgs.client.gtp.f a PLUS FIN: Got challenge from agotaoxin, testing engine response. 30 déc. 2011 01:32:20 com.gokgs.client.gtp.GtpClient d LE PLUS FIN: Command sent to engine: boardsize 19 30 déc. 2011 01:32:20 com.gokgs.client.gtp.GtpClient d LE PLUS FIN: Got successful response to command boardsize 19: = 30 déc. 2011 01:32:20 com.gokgs.client.gtp.f a LE PLUS FIN: Board size 19 is acceptable 30 déc. 2011 01:32:20 com.gokgs.client.gtp.GtpClient d LE PLUS FIN: Command sent to engine: time_settings 135 0 0 30 déc. 2011 01:32:20 com.gokgs.client.gtp.GtpClient d LE PLUS FIN: Got successful response to command time_settings 135 0 0: = 30 déc. 2011 01:32:20 com.gokgs.client.gtp.f a LE PLUS FIN: Time system is acceptable 30 déc. 2011 01:32:20 com.gokgs.client.gtp.f a LE PLUS FIN: Sending challenge back to server 30 déc. 2011 01:32:21 com.gokgs.client.gtp.GtpClient e GRAVE: Unexpected disconnect: The connection to the server has been closed unexpectedly. You must log in again. 30 déc. 2011 01:32:21 com.gokgs.client.gtp.GtpClient go FIN: Will wait 5 minutes, then try to connect again. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
Thanks Jouni for buying Crazy Stone. And yes, Don, there is an Android version. All the many versions of Crazy Stone are listed on its web page: http://remi.coulom.free.fr/CrazyStone/ Chinese versions coming soon. Rémi On 10 janv. 2012, at 23:48, Don Dailey wrote: Is there an Android version for my tablet? Don ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
Hi Vlad, Sorry for your medal. I don't do the UI stuff, but I'll forward your remark to Unbalance. Or if you can send the position with the wrong scoring to me, I'll try to make CS score it correctly. Rémi On 11 janv. 2012, at 15:24, Vlad Dumitrescu wrote: Hi, On Wed, Jan 11, 2012 at 15:07, Don Dailey dailey@gmail.com wrote: Thanks, I found it on the Android Market and purchased my copy. I just did that too (especially since it's on sale at only €2,99), and one thing that I'm not happy about is that when a game is finished, there is no way to contest the scoring done by the program. It's annoying when one doesn't get the medal one deserves because a dead group was marked as alive :-) regards, Vlad ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
I wonder what kind of abuse you censor. I never observed any behaviour I'd like to censor with Crazy Stone. Rémi On 10 janv. 2012, at 21:19, Jean-loup Gailly wrote: I know the Zen author occasionally logs in as the bot and censors people who abuse the bot. I do this too for pachi2. But it's a real pain to maintain the censor list. My list currently has 147 accounts (probably representing much fewer distinct persons). Jean-loup ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
On 7 janv. 2012, at 21:07, Jouni Valkonen wrote: 6d for MC-gobot seems a bit optimistic, I would bet €5 that it won't happen in 2012. 6d's are ridiculously strong. I calculated that in January, out of 56 games against 5d's CS won 48%. This is indeed impressive, although there is long way to beat 6d's in even game. I take the bet. Zen19D is 5.99 dan today. No way it won't improve to 6d. I would take the bet for Crazy Stone too. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
On 8 janv. 2012, at 16:25, Thomas Wolf wrote: Reaching 6 Dan may well happen, I only want to say that the difference between 5.99 dan and a stable 6.00 dan can be big because then all players get one more stone, all games are somewhat different. This all depends on the rating system. Depending on how well the rating system is calibrated, reaching 6d might make things easier or harder. I expect that for computers, reaching 6d will make things easier, because the level of play of computers is very uneven. Some of the fast progress of Crazy Stone when it passed the 5d bar might come from that phenomenon. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CLOP and BOOL Parameters
You can have a bounded integer parameter. For instance: IntegerParameter my_parameter 0 1 CLOP may not be the most efficient approach to tuning boolean parameters, because it makes little sense to do quadratic regression over boolean parameters. But it should work. Rémi On 4 janv. 2012, at 10:34, ds wrote: Hello, I had a first look at CLOP and it seems really easy to use. Great! One question arises: Can I add bool Parameters, e.g. to enable playout features or must I change them by hand? Thanks a lot Detlef ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CLOP and BOOL Parameters
Thanks for the link. I tried to read the manual. The table of content has a
promising entry:
7.23 Maximizing or minimizing a function of several variables
But there is no such entry in the text! I'd be curious if anybody knows where
this mysterious section is hidden.
Also, gosset is limited to polynomial regression. For binary outcomes, logistic
regression is really better. But optimal design for logistic regression is a
bit complicated, and not supported by gosset. Anyway, I spent a lot of time
studying optimal design theory when developing CLOP. My feeling is that it does
not work. This is the relevant paragraph in my paper:
The next sample is chosen at random (using Gibbs sampling) by using $w$ as a
probability density. The theory of optimal design offers many alternatives
\cite{Chaloner-1989a,FackleFornius-2008a}, but sampling according to $w$
outperformed them in experiments. A problem with optimal design is that it
assumes that the model perfectly fits the data, which is always wrong in
practice. Even when lack of fit is taken into consideration~\cite{Wiens-2010a},
these algorithms take samples near the edge of the sampling domain. Samples
near the edge are rapidly discarded when the domain is shrinking. It might be
possible to do better, but sampling according to $w$ is simple and works well.
Rémi
On 4 janv. 2012, at 12:40, steve uurtamo wrote:
there exists a general tool for doing things like this:
http://www2.research.att.com/~njas/gosset/index.html
you can optimize over discrete parameters, continuous parameters,
etc., by describing them, generating the experiments, and iterating.
it's not simply an optimization tool, so requires more hands-on work,
but it's quite good.
s.
On Wed, Jan 4, 2012 at 2:42 AM, Rémi Coulom remi.cou...@free.fr wrote:
You can have a bounded integer parameter. For instance:
IntegerParameter my_parameter 0 1
CLOP may not be the most efficient approach to tuning boolean parameters,
because it makes little sense to do quadratic regression over boolean
parameters. But it should work.
Rémi
On 4 janv. 2012, at 10:34, ds wrote:
Hello,
I had a first look at CLOP and it seems really easy to use. Great!
One question arises: Can I add bool Parameters, e.g. to enable playout
features or must I change them by hand?
Thanks a lot
Detlef
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] win rate bias and CLOP
On 4 janv. 2012, at 16:22, Olivier Teytaud wrote:
BTW, I imagine that CLOP could be any fully automated parameter tuning
solution. That is, nothing here is really specific to CLOP. It just happens
that CLOP is the first fully automated parameter tuning system that I have
made to work.
Hi Brian and others;
I'm not sure that CLOP could be any fully automated parameter tuning solution.
The point is that there are roughly two families of algorithms:
1) those in which exploration is roughly equal to recommendation; we test
parameters close to our current approximation
of the optimum;
2) those in which exploration is really different from recommendation: we
test parameters possibly far from the optimum,
because the variance is higher and we get more information there.
I assume that CLOP is in category 2.
I don't think CLOP fits in any of these categories. CLOP tries to balance bias
and variance optimally by sampling over an area that is as small as possible
(to avoid regression bias because the function to be optimized might not be
quadratic or symmetric), but big enough to be confident that values on the
border are inferior to the optimal win rate (otherwise we can have no idea
where the maximum is).
The big deal with CLOP, compared to methods such as UCT or CEM, is not about
sampling far or close. The sample distribution of CLOP is very similar to the
sample distribution of Gaussian CEM. The big deal with CLOP is that it is an
order of magnitude faster when the function to be optimized is smooth (bounded
second-order derivative, say). That is because it is not a comparison-based
algorithm (like population-based methods). It seems that comparison-based
approaches cannot do better than 1/sqrt(N), whereas CLOP can do N^{-2/3}
asymptotic expected simple regret.
The function to be optimized does not have to be symmetric. I know optimizing
symmetric functions is very popular in some theoretical literature of
stochastic optimization, but I don't expect these results have much practical
value, because functions to be optimized in practice are never symmetric around
the maximum. Lack of symmetry really changes the nature of the problem.
The really fundamental result in terms of asymptotic speed of convergence is
that paper:
@article{
Chen-1988a,
author = Hung Chen,
title = Lower Rate of Convergence for Locating a Maximum of a Function,
journal = The Annals of Statistics,
volume = 16,
number = 3,
pages = 1330--1334,
year = 1988
}
in short: bounded d-th derivative = O(N^{d/(d+1)}) lower bound on the
asymptotic rate of convergence. This is a universal result. It applies to _any_
algorithm. I must admit I don't really understand the theorem, but I
intuitively agree with it.
Rémi
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
As a human player, I would be opposed to playing against a program in a tournament. So I don't wish to participate in human tournaments with Crazy Stone. Also, having to operate the program manually would be a huge pain. Playing casual games automatically on KGS is much more pleasant for me, and not less interesting to watch than tournament games. Regardless of time control, the level of play is immensely higher than mine, anyway. Rémi On 3 janv. 2012, at 13:09, Ingo Althöfer wrote: Hello, I am in charge of organizing the computer go parts of the European Go Congress 2012. Before asking the main congress organizers if they were willing to allow a computer player in the main tournament I would like to make sure that - when allowed - really a programmer with a strong bot shows up. So, programmers of strong bots: Feel free to contact me soon (best by email - I will treat them confidential) if you are willing to come to Bonn. Your machine may be remote, but you or a member of your team should be in Bonn to operate the program. Some of the strong programs that come to my mind as possible participants are (in rather random order) Zen, CrazyStone, Steenvreter (Bonn is not very far from the NL), Erica, ManyFaces, Pachi, MoGo, Aya, Gomorra. Of course, such a participation would mean that each human participant has to declare at the start of the tournament whether he/she is willing to be paired against the bot - and these declarations have to be obeyed. Also, computer participants should be excluded from prize money. Jouni wrote: ... I would love to see strong gobot playing in EGC 2012 main tournament. Of course there is lots of organizing thing to do, but those two evil gobots are good enough and still not yet too good to participate into serious human tournaments with long thinking times. EGC 2012 takes place in Bonn (Germany, Rhine area), starting on July 21, and ending on August 04, 2012. The open tournament has 10 rounds, see at: http://www.egc2012.eu/congress/tournaments/open-european-championship In case of a/one computer participant in the main tournament I am willing to help by providing a present for each human player who plays the computer. Ingo. -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
On 2 janv. 2012, at 20:12, David Fotland wrote: Very impressive. Does anyone know the relative hardware? I think the 5 dan Zen was running on 26 cores. Thanks for the nice comments. I am running on 24 cores (Dell PowerEdge R905 Rack Server, 4 x Six-Core AMD Opteron(tm) Processor 8439 SE, 2.8GHz). But it is only one machine. So my 24 cores are probably much more efficient than the 26 cores of Zen. Remi, can you tell us how much of the benefit was due to CLOP? It's an excellent tool BTW. Thank you for sharing it. The big improvement I got recently came from tuning one parameter that I had forgotten to tune for a few years. I tuned it with CLOP, but I would have probably got the same result if I had tuned it manually (about +100 Elo in self play, both on 9x9 and 19x19). CLOP simply makes tuning a lot more convenient. I really enjoy using it very much. I am glad if people like it too. I also fixed a silly bug that was worth 30 Elo in self-play. Improving the KGS rank at 5d is very difficult. The current experimental version of Crazy Stone wins 78% against Crazy Stone 2011. That's less than one stone difference on KGS. But I am sure both Zen and Crazy Stone will reach 6d in 2012. Rémi -David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Ingo Althöfer Sent: Monday, January 02, 2012 11:04 AM To: computer-go@dvandva.org Subject: [Computer-go] CrazyStone in the 5-dan footsteps of Zen Hello everybody, let us start the new year also in the mailing list. It seems that Zen is finding a strong contender for its position as leading bot on KGS. Starting on December 29, Crazy Stone (by Remi Coulom) has started a long playing session in KGS computer room, and has now established a stable 5-dan rating. See the rating diagram for CrazyStone at http://www.gokgs.com/graphPage.jsp?user=crazystone and the list of games at http://www.gokgs.com/gameArchives.jsp?user=crazystoneyear=2011month=12 and http://www.gokgs.com/gameArchives.jsp?user=crazystoneyear=2012month=1 Ingo (wishes to see bots with 6-dan ranks on KGS before the European Go Congress in July 2012) -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Results of first day of the UEC Cup
Hi, A video with the results of the first day is online there: http://www.ustream.tv/channel/uec-go-2011 If I understood correctly, the top players may be: 1. Zen 2. Erica 3. Aya 4. Pachi 5. Many Faces 6. Nomitan But my japanese is not so good :-) Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] ACG13 live broadcast and programme
There will be a live video broadcast of the ACG13 conference: http://www.grappa.univ-lille3.fr/icga/news_item.php?id=71 This a link to the programme: https://www.conftool.net/acg13/sessions.php Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Question about exploration in CLOP
My implementation is very basic (and inefficient). I use Gibbs sampling (ie, Metropolis-Hastings, one dimension at a time, which scales better to higher dimensions), with uniform samples over the parameter range. Details of the implementation are in CSPWeight.cpp. I found it is good enough in practice, but I will improve it. It should be easy to use the quadratic regression to define a candidate distribution that is much better than uniform. http://en.wikipedia.org/wiki/Metropolis%E2%80%93Hastings_algorithm http://en.wikipedia.org/wiki/Gibbs_sampling Also, an unrelated note about priors: it is a good idea to use a pessimistic prior for the mean/lcb estimation, and a more optimistic prior for the regression. I did not mention it in the paper. It prevents the algorithm from iterating forever until the winning rate is close to 100%. It is not extremely critical for performance, but it may help a bit. Rémi On 15 nov. 2011, at 17:59, Brian Sheppard wrote: I would like to know more about the exploration methods that you tested in CLOP. Let's start with Metropolis-Hastings. I understand Metropolis-Hastings as having a current point P, which has a weight Wp, and randomly sampling a point Q, which has weight Wq. Then your next point will be Q if Wq = Wp, or if Wq Wp then move to Q with probability Wq/Wp, and keep P otherwise. Do I have that right? My question concerns the space over which Q is sampled. Is it just random over the whole domain? Or a radius around P? Thanks, Brian ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Possible approach to prove CLOP convergence?
I have no idea. I did not really spend any time trying to prove mathematically that CLOP works. Another source of ideas might be that paper about a similar method, where they propose a proof of convergence: http://www.informs-sim.org/wsc07papers/041.pdf But the algorithm is extremely complicated, and I did not understand that proof (they don't give details, only theorems). I wonder how they can have theorems that have no probabilities or expectations in them, when they have noisy measurements. I expect any convergence proof should be a proof of stochastic convergence (ie, almost sure, with probability one, ...). Rémi On 9 nov. 2011, at 01:34, Brian Sheppard wrote: Is it possible that convergence properties of CLOP can be established by relating it to Boosted Regression? The successive exponential weights remind me of boosting... Brian ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] New version of CLOP
The prior strength was increased from 1e-3 to 1e-2, and the number of iterations of the localization process was limited to 7. I did not have time to measure this very precisely, but tests on Rosenbrock5 show a performance improvement. I made that change because some users of CLOP sent data to me that show that when CLOP had many parameters (10), it had a tendency to overfit: the localization iteration looped until only a few wins were left. That did not prevent the algorithm from working, but it was a bit inefficient. Rémi On 7 nov. 2011, at 03:50, Brian Sheppard wrote: How does the new version improve regularization in high dimensions? My diff of the source code showed some changes, but it was not clear which changes improve regularization, specifically in high dimensions. Thanks, Brian ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] New version of CLOP
On 7 nov. 2011, at 14:07, Brian Sheppard wrote: I have also seen runs where many localization iterations are necessary, but I did not connect that to overfitting. I do see Clop runs that wander for an excessive number of trials. That has troubled me, because I would like to use Clop for high dimensional tuning, but I cannot afford to run 1e7 trials. Apart from hardware availability, there are two objections to really long runs: - I am sure to change my program before it finishes the run. - It might be worse than doing 1e3 runs of 1e4 trials each, using smaller parameter families. A lot depends on the correlations between parameters. For example, if parameters are independent then the potential benefit of tuning all concurrently is minimized. If the parameters are intricately related, then there could be great benefit in concurrent tuning, since tuning one parameter at a time might not see the big picture. I am curious: can Clop exploit a covariance matrix to factor the samples into minimally interdependent sets? For example, in the case of fully independent parameters, then Clop could choose x0 by ignoring all coordinates but x0 in its sample. The challenge is to create a scalable form of dimensionality reduction. Brian My plan is to investigate sparse methods in the future. That paper seems a good starting point: http://en.scientificcommons.org/43266410 But I don't expect it can do miracles. Allowing the user of CLOP to manually indicate groups of parameters that are expected to be correlated might be another approach. I tried CLOP in the completely independent case (the Log5 problem) with independent quadratic regression (ie, all non-diagonal terms of the hessian are set to zero), and full quadratic regression. The results was presented in that CCC message: http://www.talkchess.com/forum/viewtopic.php?topic_view=threadsp=423257t=40237 So, independent regression is not a big win. Right now, one major problem for using CLOP in high dimensions is the computational cost. That's because my implementation of quadratic logistic regression is not efficient. It is based on Newton's method. I will try conjugate gradient in the future, as it seems to be a much more efficient approach to logistic regression in high dimensions. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] New version of CLOP
Hi, I have just uploaded a new version of CLOP. Changes: 2011-11-05: 0.0.9 - Stronger regularization (avoid overfitting in high dimensions) - Merge Replications option in gui - faster, better display - Performance optimization of display and loading of large data files - Removed -ansi option for Windows compilation - Shrinking parameter ranges does not lose integer data any more - Removed confusing columns: max-1, max-2, ... - More explanations in the doc: biased win rate + GammaParameter You'll also find new slides of a seminar presentation I made this week at the University of Essex, as well as Windows binaries. http://remi.coulom.free.fr/CLOP/ Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] MCTS playouts per second
I think it is more like 100k on 9x9, and 25k on 19x19 http://www.mail-archive.com/computer-go@computer-go.org/msg11214.html Rémi On 28 oct. 2011, at 06:30, Aja Huang wrote: No, I meant 6,000-7,000 playouts per second on 19x19. Aja From: Michael Williams Sent: Thursday, October 27, 2011 5:18 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] MCTS playouts per second Perhaps you meant to say 60,000-70,000 playouts per second for libego? On Wed, Oct 26, 2011 at 10:23 PM, Aja Huang ajahu...@gmail.com wrote: On 19x19, Erica's speed is around 5,500 lightweight playouts per second on a single i7 cpu. As far as I know, Lukasz Lew's libego, which is open source, is the fastest implementation of MCTS and can reach around 6,000-7,000 lightweight playouts per second in the same cpu. Aja -原始郵件- From: Scott Christensen Sent: Wednesday, October 26, 2011 6:48 AM To: computer-go@dvandva.org Subject: [Computer-go] MCTS playouts per second Just want to check what the expected playout performance is of well tuned monte-carlo engines? My MCTS engine is averaging apx 3,500 lightweight playouts per second on a single i5 32 bit cpu. Any suggestions on very efficient source code examples for fast monte-carlo playouts? I've spent a lot of time comparing recursive group formation vs non-recursive but it doesn't seem to make a big difference. It seems that updating the list of likely moves after every play with something similar to the mogo probability rules is the most time consuming part as I currently recalculate the probabilities of moves at every empty point on the board each turn. It seems necessary if one doesn't want to handle all the exceptions to keeping the previous turn's play probabilities. Also any thoughts on combining pattern scoring and other conventional techniques together with a UCT tree? If two branches have very similar simulated win ratios could one use other factors to choose the best branch? It seems if there is a very wide branching such as at the beginning of the game, there is a lot of room to add other heuristics to choosing the best move when monte-carlo scores are within range of expected error. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CLOP: Confident Local Optimization forNoisyBlack-Box Parameter Tuning
Hi Brian, On 4 oct. 2011, at 18:54, Brian Sheppard wrote: Hi, Remi. I have a question about the burn-in process for CLOP. Normally you need a lot of data to make a decent regression function. For example, if you have N arguments in your function, then CLOP (Correlated-All) needs 1 + N * (N+3) / 2 parameters. So if you want 10 observations per parameter, then you need 10 + 5N(N+3) samples. But even getting *one* sample can be tricky, because the 'logit' for a sample is +INF if the sample wins all of its games, and -INF if the sample loses all of its games. So you need a sample that has some wins and some losses. If the true value of the function is near 0.5, then the average number of trials required to obtain a sample is around 3, which is fine. I deal with +INF/-INF with a prior: the Gaussian prior regularizes the regression, so its tends to remain flat and close to 0.5 when very few samples have been collected. But some of the test functions in your paper are very different. For example, the Correlated2 function is nearly 0 for most of the domain [-1,1]^4. When I sample randomly, it takes ~5K samples (that is, ~20K trials) to turn up enough samples to fit a regression line. I am not sure I understand what you mean. If you use regularization, you can perform regression even with zero samples. Of course, it is very inaccurate. But if you are careful to take confidence intervals into consideration, you can still do statistics with very few samples, and determine with some significance that an area is bad. I tried initializing my win/loss counters to epsilon instead of zero. But that technique was not robust, because any reasonable epsilon is actually larger than Correlated2 for most of the domain. Consequently, the reduce the weights step does not reduce enough weights, and the logistic regression ends up fitting epsilon, rather than Correlated2. So I cannot get a valid measurement with less than 20K trials before the first regression step. But your paper shows regret curves that start out at 10 trials. What am I missing? I am not sure what you are missing. In the case of Correlated2: In the beginning CLOP will sample uniformly at random (if you run the algorithm in the paper with N=0, then w(x)=1 everywhere). As soon at it find its first win, it will start focusing around that first win. You should be able to easily run CLOP on Correlated2. Just edit DummyExperiment.clop and DummyScript.py. You can also take a look at Gian-Carlo's chess data: it is a bit similar, as most games are lost in the beginning. One important aspect of CLOP is the use of the confidence interval. It does not matter if the regression is very inaccurate. Even with an inaccurate regression, it can get confident that some areas of the search space are below average, so they should not be sampled. If you sample uniformly at random until you get an accurate regression, then, yes, it will take forever. Maybe what you are missing is that CLOP does not need an accurate regression at all to already focus its sampling on a promising region. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Computer Olympiad: Go schedule and registration cost for multiple tournaments
Hi, I am glad to report that because of remarks about registration cost, the organizers have agreed to reduce the cost of additional tournaments to 25 Euros for the game of Go if you already have registered for one Go tournament. So, first Go tournament is normal price, and each additional tournament is 25 Euros instead of 40. Rémi On 27 sept. 2011, at 18:37, Petr Baudis wrote: Hi! On Thu, Sep 22, 2011 at 07:28:57PM +0200, Rémi Coulom wrote: I have just posted this on the web site of the Olympiad: Schedule for Go tournaments: • 21-22 nov go 9x9 • 23 nov go 13x13 • 24-25 nov go 19 x 19 • 26 nov. 19x19 play-off if any, or a demonstration game Participants have to register separately for each tournament. Participants who wish to participate in the Olympiad don't have to pay a full registration for each tournament. One full registration has to be paid, plus 40 Euros for each additional tournament, regardless of category (see details in the registration form). http://www.grappa.univ-lille3.fr/icga/news_item.php?id=67 When researching the conference at first, I originally thought that a conference fee is also enough for the tournament, but I got confused when reviewing this now. So just to clarify, when I want to attend the conference and enter Pachi for 9x9 and 19x19, I will have to pay: 120 EUR student conf. fee + 80 EUR tourn. fee + 40 EUR tourn. fee? Do I understand that right? I probably won't find enough money for that, especially as 9x9 comes first (I would consider paying extra 80 EUR for 19x19, IMO is nowadays much more interesting, but paying for hotel and staying abroad longer is just not worth it for me). On a related note, anyone here attending the conference looking for a hotel roommate? Or is there a better venue for finding one? Thanks, Petr Pasky Baudis ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Computer Olympiad: Go schedule and registration cost for multiple tournaments
On 24 sept. 2011, at 17:06, Erik van der Werf wrote: BTW Are you planning to participate with Crazystone? I will attend the conference, and present the CLOP paper there, but I don't plan to participate in tournaments. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Computer Olympiad: Go schedule and registration cost for multiple tournaments
On 23 sept. 2011, at 02:53, Hideki Kato wrote: Rémi Coulom: de4c2e5c-94d2-4ef0-9186-8738a6690...@free.fr: Hi, I have just posted this on the web site of the Olympiad: Schedule for Go tournaments: • 21-22 nov go 9x9 • 23 nov go 13x13 Thank you Rémi, but 13x13 Go is not in the tournaments list on the web page (http://www.grappa.univ-lille3.fr/icga/event.php?id=43). The list of tournaments is not complete. I added it, and all the games of last year. The general idea is that the ICGA is willing to add any game to the Olympiad, provided that enough competitors register. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Computer Olympiad: Go schedule and registration cost for multiple tournaments
Hi, I have just posted this on the web site of the Olympiad: Schedule for Go tournaments: • 21-22 nov go 9x9 • 23 nov go 13x13 • 24-25 nov go 19 x 19 • 26 nov. 19x19 play-off if any, or a demonstration game Participants have to register separately for each tournament. Participants who wish to participate in the Olympiad don't have to pay a full registration for each tournament. One full registration has to be paid, plus 40 Euros for each additional tournament, regardless of category (see details in the registration form). http://www.grappa.univ-lille3.fr/icga/news_item.php?id=67 Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
