Re: [Computer-go] Aya won December 2014 KGS bot tournament
On Mon, Dec 8, 2014 at 2:04 PM, Aja Huang ajahu...@gmail.com wrote: I hope this list is not dying. :) :) It's still broken. The majority of the messages I see in the archive never arrive in my gmail account. Google seems to make some exceptions for emails from other gmail users, which suggests that somehow they are actively filtering even my spam box! Another weird thing, perhaps related, is that nearly all the spam that still gets through (to my spam box) is in Chinese. For compgo at least I can check the archive, but I wonder how many other legitimate messages I may have missed which I will never know about... On Mon, Dec 8, 2014 at 11:02 PM, Rémi Coulom remi.cou...@free.fr wrote: 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/ Yesterday the archives at dvandva.org were also down for a while. Not sure when they came back up, but Hiroshi might have run into the same issue. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Computer-Go list seems to be dying? Time to move?
On Thu, Dec 4, 2014 at 12:49 AM, Petr Baudis pa...@ucw.cz wrote: Hi! On Thu, Dec 04, 2014 at 12:12:52AM +0100, Erik van der Werf wrote: I just noticed in the archives that none of the last 3 messages sent to this list reached my gmail account. One of them was by a well known conference spammer, but the other two (by Remi and Ingo) definitely should have made it through. Like Ingo, I recently also tried to register another address but that didn't work either. Perhaps we need to move the list elsewhere? I run a mailman instance that handles various mailing lists - Pachi's, but also e.g. a fairly lively mailing list for Czech beekeepers (really!) and I do try to take care of keeping delivery from that host to other mailservers working. I can host this mailing list, ideally even back in the computer-go.org domain (otherwise, as computer...@v.or.cz). Michael, what do you think? Would you be ok with me hosting the mailing list? Or do you plan to take a look at the delivery problems yet? Sound great Petr! I was considering to offer hosting it myself, but my host doesn't have mailman installed (it has majordomo), and I really like the archive functionality we currently have as provided by mailman (essentially it's the only thing that still seems to work reliably). @Álvaro: the problem for me is not so much that messages are labeled spam (your solution will solve that); the problem is that from the current host some messages simply don't arrive at all (so not even in my spam folder). For instance, I could only read your reply by checking the archive. @Dave: Yes I understand it requires some maintenance to keep things in shape, but we are currently in a situation where a substantial part of the messages never arrive and new people cannot even subscribe to the list (at least me and Ingo weren't able to register a new address recently). So unless Michael or one of us takes action this list seems destined to die. Best, Erik PS I BCCed my original message to Michael on both his dvandva and yahoo address, but so far didn't see a reply. Also on an earlier message I never got a reaction. Does anyone know if he's ok? ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Computer-Go list seems to be dying? Time to move?
I just noticed in the archives that none of the last 3 messages sent to this list reached my gmail account. One of them was by a well known conference spammer, but the other two (by Remi and Ingo) definitely should have made it through. Like Ingo, I recently also tried to register another address but that didn't work either. Perhaps we need to move the list elsewhere? Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to AyaMC!
Hi Hiroshi, Why do you call it a static bonus? If n increases with the number of simulations, the effect of the bonus term still fades away. Perhaps the interesting part is in fading out more slowly than with ordinary priors (i.e., by 1/sqrt(n) instead of 1/n)? BR, Erik BTW Nicks original message never arrived in my gmail account :-( On Sat, Nov 22, 2014 at 6:53 PM, Hiroshi Yamashita y...@bd.mbn.or.jp 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] K-best mode in MCTS ?
I have the exact same issue, and it is not just with the Computer Go mailing list. Gmail is becoming more and more unreliable when it comes to mailing lists (including one I host myself). The most annoying aspect is that some messages don't arrive at all (not even in a spam box). Maybe it's time to ditch the likes of google and gmx. Any suggestions for a better alternative? Erik On Sun, Nov 16, 2014 at 7:21 PM, Alexander Kozlovsky alexander.kozlov...@gmail.com wrote: On Sun, Nov 16, 2014 at 6:08 PM, Ingo Althöfer 3-hirn-ver...@gmx.de wrote: It is a pity that the mailing list eems to work from an IP account that is not trusted my German mail provider gmx (unrelated to google). So, I do not get the mails, but insdtead have to look them up in the archives. Do other participants face the same problem? Gmail marked computer-go mails as spam, but I added appropriate Gmail filter with the action never send to spam. This way I received your last message. But even with that filter I never received previous messages with the topic K-best mode in MCTS. ___ 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
Well, not a laptop, but a '6-core desktop' can still be quite fast... E.g., the 48-core amd machine I often use is typically only around 2.5 times faster than a stock i7 3930k (and then I haven't even considered the i7's overclocking potential). IIRC Remi usually uses a 24-core amd machine, so I doubt the difference would be huge. Perhaps Remi can say what exactly was under the hood? On Sun, Oct 5, 2014 at 7:09 PM, Xavier Combelle xavier.combe...@gmail.com 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] Interviews on codecentric Challenge
Thanks Detlef, that makes sense. the 5960 is an 8-core (and ridiculously over-priced IMO). I actually got an 5820k this week. It should only be slightly slower than the 5930k, but it is still a bit faster than that 3930k I mentioned before. On Sun, Oct 5, 2014 at 8:09 PM, Detlef Schmicker d...@physik.de wrote: Sorry 5930x Am Sonntag, den 05.10.2014, 19:48 +0200 schrieb Erik van der Werf: Well, not a laptop, but a '6-core desktop' can still be quite fast... E.g., the 48-core amd machine I often use is typically only around 2.5 times faster than a stock i7 3930k (and then I haven't even considered the i7's overclocking potential). IIRC Remi usually uses a 24-core amd machine, so I doubt the difference would be huge. Perhaps Remi can say what exactly was under the hood? On Sun, Oct 5, 2014 at 7:09 PM, Xavier Combelle xavier.combe...@gmail.com 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 ___ 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] Komi for 13x13
Here's an intuitive argument for why I believe 7.5 komi is too high on any of the larger boards. On very small (square odd-surface) boards there is no room for White to live (Black takes all points). Past a certain point (5x5), Black can no longer control the full board, so White also starts to take some points. Initially there is not much room left for White, so the perfect komi remains a bit high (e.g., 9 for 7x7), but as the size increases the White move values start to approach those of Black and I expect the advantage of the initiative to decrease to a constant (probably 7 or 5 points). Since we have reasonable indications that the first player score on 9x9 is no higher than 7, and because it is unlikely to increase with size, the main question is if it remains at 7, or if it drops down further (e.g., to 5). (even values such as 6 are much less likely on odd-surface boards with area scoring). Of course one could still argue that 7 is correct, and we just add 0.5 to prevent ties, but then I'd rather subtract 0.5 and go for a komi of 6.5 because I prefer the small advantage to go to the first player. Best, Erik BTW Nick, is there any chance that you will at some point run a kgs tournament with territory scoring? I'm still hoping that one day we will play Japanese rules with 6.5 komi. (Regardless of the komi, I think the more fine-grained territory scores would make the games more interesting, especially on 9x9.) On Tue, Aug 5, 2014 at 11:10 AM, Nick Wedd n...@maproom.co.uk wrote: In January 2011, I took a poll on what komi to use for 9x9 bot tournaments on KGS. There was a majority in favour of changing from 7.5 to 7. During Sunday's 13x13 KGS bot tour, someone suggested also changing the 13x13 komi from 7.5 to 7. The main argument for changing must be that 7.5 favours White. So here are the statistics from recent KGS 13x13 bot tournaments: White Black wins wins 2014 August 13 9 April 36 27 2103 December12 18 April 28 20 TOTALS 89 74 I am inclined to think there is not a convincing reason to change. But I will welcome persuasion; or better, more statistics. 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] Komi for 13x13
On Tue, Aug 5, 2014 at 2:20 PM, Nick Wedd n...@maproom.co.uk wrote: On 05/08/2014 12:25, Erik van der Werf wrote: Here's an intuitive argument for why I believe 7.5 komi is too high on any of the larger boards. On very small (square odd-surface) boards there is no room for White to live (Black takes all points). Past a certain point (5x5), Black can no longer control the full board, so White also starts to take some points. Initially there is not much room left for White, so the perfect komi remains a bit high (e.g., 9 for 7x7), but as the size increases the White move values start to approach those of Black and I expect the advantage of the initiative to decrease to a constant (probably 7 or 5 points). Since we have reasonable indications that the first player score on 9x9 is no higher than 7, and because it is unlikely to increase with size, the main question is if it remains at 7, or if it drops down further (e.g., to 5). (even values such as 6 are much less likely on odd-surface boards with area scoring). Of course one could still argue that 7 is correct, and we just add 0.5 to prevent ties, but then I'd rather subtract 0.5 and go for a komi of 6.5 because I prefer the small advantage to go to the first player. I find your logic convincing. We'll see if others do. But even if they do, is that a sufficient reason for changing the komi in events that I run? With 9x9, getting the komi slightly wrong biases the results markedly, and so should be avoided. With 19x19, getting the komi slightly wrong makes much less difference to the results of games, and it may be better to use slightly wrong komi than to inconvenience those bot authors who have not persuaded their bots to understand integer komi and the possibility of jigo. Sure, I understand. But on the other hand, does it really matter if occasionally some bot doesn't get jigo? Also, they have to get this right anyway for the 9x9 tournaments. I think the choice should be made based on weighting the advantage of having no ties (always a winner, easy pairing) against the amount of bias in the starting condition (so it also matters if, e.g., you run single or double round robin). Weighting the two, I prefer fractional komi for 19x19 and integer komi for 9x9. For 13x13 I'm not sure, but I'm tempted to go for integer komi as well. In any case, if you don't want integer komi you could still switch to 6.5. It might be interesting to see if that turns it into hinting at a slight advantage for Black. BTW Nick, is there any chance that you will at some point run a kgs tournament with territory scoring? I'm still hoping that one day we will play Japanese rules with 6.5 komi. (Regardless of the komi, I think the more fine-grained territory scores would make the games more interesting, especially on 9x9.) I use area scoring because, with the KGS clean-up procedure, it allows almost all games to end without my intervention. KGS does not offer a similar mechanism for what it calls Japanese rules. I do not want to be required to apply my rules knowledge and counting ability in a game which ends with the players disputing the status of a bent four in the corner. Ah yes, that makes sense. Implementing a proper cleanup procedure for territory scoring is not difficult. Of course a practical implementation might differ slightly from the official Japanese rules, but then again KGS didn't really implement the official Chinese rules either... Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] What happened with the August bot tournament
On Fri, Jul 25, 2014 at 7:44 PM, Nick Wedd n...@maproom.co.uk wrote: ... If there are less than seven entrants, the usual Swiss tournament will be replaced by a double-round-robin tournament (each player will play each other players twice), starting five minutes later, at 08:05. This tournament's details are at http://www.gokgs.com/tournInfo.jsp?id=912 . On Sun, Aug 3, 2014 at 12:21 PM, Ingo Althöfer 3-hirn-ver...@gmx.de wrote: Hi, I just looked into the website of the August bot tournament. Big surprise: no more 7 or more applicants, but only two unknown players. What has happened? 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] Congratulations to Zen!
I think in cases like this it would be better to reduce to (double) round-robin. Erik On Mon, Jul 7, 2014 at 6:26 PM, Nick Wedd n...@maproom.co.uk wrote: Congratulations to Zen19S, winner of the July KGS computer Go tournament, with seven wins from seven games! My very short report is at http://www.weddslist.com/kgs/past/104/index.html As usual I will welcome any 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] C++11; threads
2k/s doesn't sound too bad (if it's a rather heavy playout policy), but I would expect a lot more than 200 moves for 19x19. On my phone I only get a few hundred 19x19 playouts per thread per second... Erik On Thu, Jun 19, 2014 at 12:15 PM, Chun Sun sunchu...@gmail.com wrote: Thanks David, I ask this because I can only get 2k/s on 19x19 with 200 move per playout, single thread, no playout termination check, and I saw people on this forum usually have much better numbers. Looks like I need to work on my performance more :) Thanks, Chun From the empty board until the playout terminates. I think for 9x9 it’s typically around 110 moves. David From: computer-go-boun...@dvandva.org [mailto:computer-go-boun...@dvandva.org] On Behalf Of Chun Sun Sent: Wednesday, June 18, 2014 8:21 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] C++11; threads Hi all, Sorry to ask this beginner question in this thread: When you say playouts per second, how many moves does each playout have? on average? Do you play from empty until the board is full for each playout? Thank you, Chun ___ 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] Bitwise-parallel surround capture
Hi Cameron, This is not new; I've been using algorithms like this in my Go programs for well over a decade. Anyone with some practical experience in binary image processing should know basic operations for dilation, erosion, growing objects under a mask, etc. Migos and Magog used a bitboard representation that did not track liberties. 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. BTW when incrementally updating the board (per move) you can stop early when the last played stone and directly adjacent opponents (if any) are found to be alive. If there's a capture, you don't need to check suicide. Best, Erik On Sat, May 24, 2014 at 11:44 AM, 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] Computer-go Digest, Vol 52, Issue 20
There are some differences in the actual implementation (e.g., I usually encode one row per int/short/byte, so vertical adjacency is by index in an array and horizontal adjacency by bit-shift), but the general idea is the same. My code is not open source and I don't directly know good references on basic mathematical morphology by heart (most of my knowledge is from nearly 20 year old university course material). However if you just look up literature on mathematical morphology for image processing I think there should be plenty. Perhaps others can pitch in; I'm fairly sure I wasn't the first to use a fast bitboard representation for Go. Erik On Sat, May 24, 2014 at 3:03 PM, Cameron Browne cameron.bro...@btinternet.com wrote: From: Erik van der Werf erikvanderw...@gmail.com This is not new; I've been using algorithms like this in my Go programs for well over a decade. Anyone with some practical experience in binary image processing should know basic operations for dilation, erosion, growing objects under a mask, etc. Hi Erik, Thanks for the info. Were they algorithms like this or this actual algorithm? Do you have any links to literature or code I can follow up? Thanks, 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] C++11; threads
This is not accurate. In my experience you should expect a substantial performance increase from hyperthreading. (For my program on an i7-3930 it was something like a 40%, Zen got a similar number, others on this list have claimed even higher numbers, e.g., see: http://dvandva.org/pipermail/computer-go/2012-August/005298.html). Erik On Mon, May 12, 2014 at 5:06 PM, Mikko Aarnos mikko.aar...@kolumbus.fi wrote: CPU cores are meant to be used by a single thread only. You can use more, but this rests on the assumption that two(or more) threads can effectively utilize a single core without too much competition over resources. This assumption is true in most situations, e.g. when we have to wait for IO or server queries or things like that often, and in these cases using HT can give a small performance boost. Now, here we are only utilizing the CPU. In this case the threads are only getting into each other's way. The effects of this can be seen very clearly with your program. It seems to scale perfectly, or very nearly so, as long as you don't use more threads than you have actual cores on your computer. When you go above that limit the scaling goes to hell and you get no improvement at all. The only way to solve your scaling problem is to get rid of the competing threads by turning off HT. I did this and have never looked back. -Mikko Aarnos ___ 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] CALL FOR PAPERS PARTICIPATION - Conference on Informatics and its applications 2014, Malaysia
Regardless, I don't think anyone on this list is interested, so labeling it spam seems entirely appropriate. Liezelle, do you even know what this list is about? Perhaps you are confused by a programming language that is called Go? Erik On Wed, Apr 30, 2014 at 6:59 PM, Liezelle Ann Canadilla lieze...@sdiwc.info wrote: Hi Dave! This conference is definitely LEGIT. Thanks! Liezelle Isn't this spam? Dave Origineel Bericht Van : lieze...@sdiwc.info Datum : 30/04/2014 15:55 Aan : computer-go@dvandva.org Onderwerp : [Computer-go] CALL FOR PAPERS PARTICIPATION - Conference on Informatics and its applications 2014, Malaysia All registered papers will be included in SDIWC Digital Library, and in the proceedings of the conference. TITLE: The Third International Conference on Informatics Applications (ICIA2014) EVENT VENUE: Universiti Sultan Zainal Abidin (UniSZA), Kuala Terengganu, Malaysia CONFERENCE DATES: October 8-10, 2014 EVENT URL: http://sdiwc.net/conferences/2014/icia2014/ OBJECTIVE: To provide a medium for professionals, engineers, academicians, scientists, and researchers from over the world to present the result of their research activities in the field of Computer Science, Engineering and Information Technology. ICIA2014 provides opportunities for the delegates to share the knowledge, ideas, innovations and problem solving techniques. Submitted papers will be reviewed by the technical program committee of the conference. KEYWORDS: Informatics, E-Learning, Information Ethics, IInformation Content Security, Anti-cyberterrorism, Real-Time Systems and many more... SUBMISSION URL: http://sdiwc.net/conferences/2014/icia2014/openconf/openconf.php SUBMISSION DEADLINE: Oct. 17, 2014 CONTACT EMAIL: icia2...@sdiwc.net ___ 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] Congratulations to Zen!
(1) We have pretty good indications that 7.0 is optimal for 9x9. (2) There is no reason to assume perfect komi to increase with boardsize, and (3) for 19x19 (with area scoring) quite likely it should be 5.0 or 7.0 (so the slope from 9x9 to 19x19 is probably quite flat). Computers are still fairly weak on 13x13, so it's not urgent yet, but if I'd have to make an educated guess I'd say 7.0 is best for 13x13. Further, if draws are considered undesirable I'd prefer 6.5 over 7.5 because I don't like first player disadvantage. Erik On Wed, Apr 9, 2014 at 7:57 PM, Nick Wedd n...@maproom.co.uk wrote: On 09/04/2014 15:39, Christoph Birk wrote: On Apr 9, 2014, at 7:24 AM, uurtamo . uurt...@gmail.com wrote: I'm not sure what you're hoping to measure with that On Apr 9, 2014 7:21 AM, Christoph Birk b...@obs.carnegiescience.edu wrote: On Apr 9, 2014, at 1:23 AM, Nick Wedd n...@maproom.co.uk wrote: Adding up the results for the last four KGS 13x13 bot tournaments, all played with 7½ komi, I find that Black has won 101, White has won 112. How does this statistic look if you only include games _between_ bots that have more wins than losses? Nick posted the above statistic (B+101, W+112). I assume he wanted to support his opinion that 7 is the proper komi for 13x13. No - I wanted to show that we have insufficient evidence to claim that 7.5 is too high. It was derived from all games in the tournaments. It was derived in all games in 13x13 tournaments since April 2013, when my reports started to include counts of black and white wins, of the form Black won 28 games and White won 32. If anyone wants to go through the results of more KGS 13x13 tournaments, excluding games involving whatever they consider to be weaker programs, and report on what they find, I will be interested. I suggest they use the links at http://www.weddslist.com/kgs/past/index.html , and that they start on or after June 2010, when I started to include cross-tables in my reports. Nick I think that games between or against weaker programs should be ignored, since they are mostly noise. Christoph ___ 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
Re: [Computer-go] Japanese rule in MC
Hi Hiroshi, You should grow the tree beyond consecutive passes. Two consecutive passes don't really end the game with Japanese rules; your engine needs to be able to figure out how to defend the large territories, e.g., in case of a resumption. Best, Erik On Sun, Mar 16, 2014 at 2:25 AM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: Hi, Recently I have tried to use Japanese rule without one point safe margin. I use Erik and Aja's method. It works pretty well, but this position is still diffcult. 9X.O.. 8X.O.. 7X.O.. Black(X) to play 6...XXOO.. komi is 6.5 5...X.O... 4...X.O... 3...X.O... 2...X.O... 1...X.O... ABCDEFGHJ (;GM[1]SZ[9]RE[B+0.5]KM[6.5]RU[Japanese] ;B[di];W[fi];B[dh];W[fh];B[dg];W[fg];B[df];W[ff];B[de];W[fe] ;B[dd];W[fd];B[ed];W[gd];B[ec];W[gc];B[eb];W[gb];B[ea];W[ga]) Best move is PASS or filling share liberties(F9,F8,F7,E5,E4,E3,E2,E1). Aya can play it, but its winrate is around 0.50, far from understanding this position is easy win for B. Tentyo no Igo(commercial version of Zen) also returns around 0.50. Is there good method that can return over 0.80 winrate? I feel Do playout and modify score by number of pass is hard for B +0.5 win position. Because B has more territory about 7 pt, and W has bigger chance to pass in playout. As a result, B winrate is lower than 0.50. My code is like this. if ( last_two_move_is_pass_in_tree ) { // confirmation phase pass_w_playout = pass_b_playout = 0; playout_length = 0; } game_length = record_length + tree_legth + playout_legnth; pass_w = pass_w_record + pass_w_tree + pass_w_playout; pass_b = pass_b_record + pass_b_tree + pass_b_playout; Margin = 0; if ( (game_length1) ) { if ( first_player_is_white ) Margin = -1; else Margin = +1; } minus = (Margin + pass_w - pass_b); japanese_score = chinese_score - minus; double final_score = japanese_score - komi - handicaps; Without confirmation phase, winrate drops from 0.50 to 0.45. Aya does not grow tree after consecutive pass in tree. Maybe I need to grow? But to do that, I need to have two root positions that have same hash value, one is root, and the other is after onconsecutive pass. Steenvreter's solution http://dvandva.org/pipermail/computer-go/2010-April/000233.html Erica's solution http://computer-go.org/pipermail/computer-go/2013-February/005757.html Many Faces' solution http://dvandva.org/pipermail/computer-go/2013-February/005748.html Useful link by Fuego team http://sourceforge.net/apps/trac/fuego/wiki/JapaneseRules 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] Game records
Myriam, I don't understand; for me the sgf format works fine for storing and retrieving timing information (using the properties Remi already pointed you at). I know some sgf editors fail to use the available sgf properties correctly, but that's not a fundamental problem with sgf. Perhaps you should try game records from another source? Erik On Mon, Feb 17, 2014 at 6:15 PM, Myriam Abramson mabram...@gmail.com wrote: The sgf format gives the timestamp of when the game started, not the timestamp of every move (possibly in milliseconds) so that won't do unfortunately. Rémi Coulom remi.cou...@free.fr writes: 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 -- myriam Go Proverb: Take the cutting stone on the second line. ___ 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
IIRC KGS does not provide useful timing information for some byoyomi settings. If you use my android app (GridMaster) to record game records you should be able to extract correct timing information for all supported settings (Absolute, Canadian, Japanese, Stopwatch) even when in overtime. Best Erik On Fri, Feb 14, 2014 at 3:36 PM, Rémi Coulom remi.cou...@free.fr wrote: 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 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] GridMaster Android App
On Tue, Dec 31, 2013 at 9:43 PM, terry mcintyre terrymcint...@yahoo.com wrote: Is this available on iPad, by any chance? No, GridMaster runs on Android, so you'd need an Android tablet (not an iPad). I use Linux, so Android was the natural choice for me. I don't have plans for iOS or Windows. Best, Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] GridMaster Android App
Thanks John, I'm glad you like it! BTW It's of course very nice to get positive feedback like this, but if anyone has more critical comments, questions, feedback or suggestions for improvement, feel free to contact me as well. Best, Erik Op 26 dec. 2013 23:14 schreef John Tromp john.tr...@gmail.com: dear Erik, The app is called GridMaster, and you can download it from: https://play.google.com/store/apps/details?id=nl.tengen.gridmaster The GridMaster package includes a lite version of Steenvreter, but I tried to make it easy to add other engines. If you're looking for something like GoGui on Android, if you'd like to play Steenvreter, if you're looking for a good sgf editor, or just something that opens Kogo's more quickly, please give it a try! I just installed it. Among the half dozen go apps on my android phone, this is now my clear favorite. Thanks for a wonderful holiday gift! regards, -John ___ 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 Crazy Stone!
Does it matter? You just remove ko threaths. The problem is with unremovable ko threaths. Op 10 dec. 2013 09:33 schreef Stefan Kaitschick stefan.kaitsch...@hamburg.de: What happens with 2 bent fours? You should be able to save one of them. On Mon, Dec 9, 2013 at 10:01 PM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: In round 3 CrazyStone vs. Aya, probably Aya was expecting W to play L12 or N12, if B's corner was not 100% dead in the playouts. It seems a bug in Aya's playouts, but even if there was no bug, Aya could still reasonably expect W's L12 or N12 in the tree. Corner bent-four is not easy to handle Yesterday I turned off bent4 code for test, and forgot. Aya understands bent4 as seki. And at end of playout, bent4 shape is counted as dead. It works well, but when outside of bent4 is semeai, it does not work. Maybe when outside W lib is 5, W has to play N12? And thanks for nice look cross-table and report. I was bit surprised I got runners-up 7 times in a row this year. 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 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Playouts vs playingstrength
Ah ok, I have little experience testing against Pachi, but obviously your scaling doesn't look good. Perhaps a bug or some fundamental knowledge that is lacking. I'd say, try to also get some data at 10^5 and make sure Pachi isn't doing any kind of pondering in the background. I remember some years ago I had some cases where tuning with very low numbers of simulations seemed to hurt scaling performance. Don't remember all the details, but obviously if you always only run a few hundred simulation per move some aspects will not be tested well. BR, Erik On Sun, Nov 17, 2013 at 2:09 PM, Detlef Schmicker d...@physik.de wrote: The graph seems not to displayed nice in all e-mail programs, here is a link http://physik.de/playouts.png Am Sonntag, den 17.11.2013, 13:52 +0100 schrieb Detlef Schmicker: Sorry, as it is an logarithmic plot it looks like a limit of +150ELO. Do other experience simelar results. There is no significant improvenment between 10^3,4 and 10^4 (3000 and 1 playouts) What might be the reason? If I do the same comparison against gnugo L10 it does not seem to have such an early limit (there is an improvnment between 3000 and 1 playouts)?! What is the problem? Can such plots guide us to where to improve the program (tree vs. playout)? Something like this:) Thanks Detlef Am Sonntag, den 17.11.2013, 13:18 +0100 schrieb Erik van der Werf: On Sun, Nov 17, 2013 at 8:21 AM, Detlef Schmicker d...@physik.de wrote: What is the experts guess?! About what? Maybe compare to Don's old study at http://cgos.boardspace.net/study/index.html Best, Erik ___ 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] How many probes down the tree are necessary for a good bot?
On Fri, Nov 15, 2013 at 12:59 AM, Darren Cook dar...@dcook.org wrote: ... I'm going by the Measuring program strength thread (Aug 3013) where people are getting 50% against GnuGo with 1K to 10K playouts.) Actually, in that thread Hiroshi was saying he only needs 350 playouts, and Detlef was at 700. So it seems you're off by roughly a factor 10. Yes, I was surprised just how heavy the Aya (and Zen) playouts must be. We've almost come full loop, and soon the programs will be traditional computer go programs doing a single playout ;-) You and Peter seem to focus quite a bit on the 'heavy' playouts while it really isn't so much about that. The reason why strong programs have to consider very few moves is knowledge applied in the tree (selection phase). Simply stated, if you know a move is bad then you don't have to run simulations on it. It is a shame the developer's skill is all dead knowledge (not open source, no papers). :-( The developers aren't dead, so the knowledge isn't dead. Further, I think most (if not all) of the required knowledge is already 'out there'; you just need the skills to put things together in the right way (and figure out which things don't work -- the typical academic paper won't tell you that). Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] How many probes down the tree are necessary for a good bot?
On Fri, Nov 15, 2013 at 10:13 AM, Stefan Kaitschick stefan.kaitsch...@hamburg.de wrote: Closed sources are indeed regretable at this point. Before, it sparked a kind of bot war, and the greatest technical advances are always made at war time. No, in the last decade the greatest advances came from academic work. Around 2007 all commercial programs were completely blown away by a few good ideas from academia. Now other commercial programs have taken back the throne and in some sense we just got back to 'normal'. As far as I can see the only open source engine that has contributed substantially to progressing the state-of-the-art is GnuGo. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] How many probes down the tree are necessary for a good bot?
On Fri, Nov 15, 2013 at 11:48 AM, Ben Ellis ben.el...@softweyr.co.uk wrote: How do you know a move is truly bad? Strong players know :-) There are almost always exceptions. Or is it a case you can mostly afford to ignore low ranked moves at the cost of the occasional lost game? In some cases we can prove it, but you don't need to be perfect. Moves that are likely to be inferior may simply be sampled (a lot) less. If a move that initially looked bad really is good, eventually it can still get sampled (if you give the program enough time). Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] How many probes down the tree are necessary for a good bot?
On Wed, Nov 13, 2013 at 3:26 AM, Darren Cook dar...@dcook.org wrote: ... I'm going by the Measuring program strength thread (Aug 3013) where people are getting 50% against GnuGo with 1K to 10K playouts.) Actually, in that thread Hiroshi was saying he only needs 350 playouts, and Detlef was at 700. So it seems you're off by roughly a factor 10. (unless of course this was for 9x9 -- I'm assuming 19x19) When people report results like this it would be nice to know what version and level of Gnugo is used... On Thu, Nov 14, 2013 at 5:12 AM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: Yes. 300 playouts on 19x19 initial board is like this, move num winrate 1:d16 64, 0.437 2:q16 76, 0.447 3:d445, 0.444 4:q4 112, 0.508 5:d171, 1.000 6:r161, 1.000 7:q3 1, 1.000 Funny, with that distribution you really don't need to search the empty board at all ;-) Best, Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] dead stones / seki detection algorithm
What type of positions are you interested in? There is certainly no simple solution for non-final positions (which is why Go is an interesting game), and even in final positions it can be tricky. For final positions Monte-Carlo playouts (with sufficient knowledge) can get very close to perfect, but there are other approaches that also work well (e.g., you could train a neural net on labeled examples). Perhaps start with Benson's algorithm for unconditional life and extend from that with whatever clever tricks you can find. Good luck, Erik On Thu, Nov 14, 2013 at 11:29 PM, Marek Jasovsky jasovsky.ma...@gmail.com wrote: Hello can someone recommend me any good and to some extent simple algorithm, I can use in software of my own, to detect dead stones and evtl seki situations, where both group remain living , as well as counting territory? thank you in advance, Marek ___ 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] On Semeai Detection
On Wed, Oct 9, 2013 at 8:54 PM, Ingo Althöfer 3-hirn-ver...@gmx.de wrote: Ps. Sorry for writing this as a new message. But for months already I do not receive the mails from the list, and an attempt to register from another mailing address failed. You're not the only one with email problems. Last month I didn't get Don's goodbye message (but saw it in the archive). Then later I didn't get the follow-up replies from Don and GCP... Most emails still seem to get through to me, but it's a pity that this list is becoming unreliable. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] checking out
Accidentally by looking at the archives I noticed that Don's last message: http://dvandva.org/pipermail/computer-go/2013-September/006252.html somehow never arrived at my gmail inbox or even its spam folder! So far this is the only message I'm aware of that failed to reach me on gmail (though an increasing number is getting labelled as spam). Perhaps there is an issue with the domain hosting the mailing list? BTW what happened to computer-go.org? In any case, I wish you all the best Don. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] List and if available site for OSS Go Engines
Fuego, Pachi and Oakfoam come to mind. Have a look here: http://senseis.xmp.net/?GoPlayingPrograms Erik On Sun, Sep 29, 2013 at 9:33 AM, Joshua Shriver jshri...@gmail.com wrote: I'm doing some testing, and would like to get and run as many Go engines as possible. Willing to dedicated 1 core and 2gigs of RAM per engine. Know there is Gnugo, but so far that is the only one I know. I can use windows or Linux. Any help is appreciated! -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
Re: [Computer-go] Japanese byoyomi in SGF files (possible KGS Bug?)
Well, I wouldn't go so far as to call it a bug in sgf, more something that just needs a better description. It is unclear what exactly 'after the move was made' means. Perhaps it should state something like 'directly after the move was made, before making any further adjustments to prepare for the next move (e.g., when in Japanese byoyomi)' I don't think there's a problem with overloading the meaning of OB/OW to handle both Japanese periods left and Canadian moves left. The OT property allows us to distinguish the two. My main concern is with the BL/WL property (not OB/OW), which for the KGS implementation of Japanese byoyomi does not provide the amount of time that was used to think about the move. One possible way to fix this would be to use both BL and WL in every node so that for every move you can provide information about the time both before and after the move is made. But if one prefers to store only information of one side (as KGS currently does), i.e. pairing BL only with B and WL only with W, then I would say that for Japanese byoyomi the time left shown should be the one at the end of the move before resetting the time for the next move. The time left after resetting is redundant anyway because the OT property already tells us the length of each period (and the OB/OW property tells use when we are in byoyomi). Best, Erik On Mon, Jul 29, 2013 at 3:50 AM, Chaz G. chaz.gwen...@gmail.com wrote: Hi Erik, I would have expected BL[2]OB[0] more than BL[2]OB[3]. Ultimately, it's a bug in sgf, not KGS. I think that BL (WL) and OB (OW) were documented with Canadian byo-yomi in mind, where BL[120]OB[3] meant that Black has 120 seconds to play 3 stones. The length, format, and number of Byo-yomi periods in a game should be tagged under OT. With Japanese byo-yomi, the literal interpretation of sgf properties would store BL[timeremaining]OB[0]WL[timeremaining]OW[0] for all periods of byo-yomi play. Using OB(OW) to mark the number of byo-yomi periods left is not the documented interpretation of sgf files but allows for a more plausible recreation of the game. Best, -Chaz On Sun, Jul 28, 2013 at 8:42 AM, Erik van der Werf erikvanderw...@gmail.com wrote: Hi All, I was just looking at how Japanese byoyomi time information is stored in sgf files. The spec at http://www.red-bean.com/sgf/properties.html tells us: Property: BL (WL) Function: Time left for black (white), after the move was made. Value is given in seconds. Property: OB (OW) Function: Number of black (white) moves left (after the move of this node was played) to play in this byo-yomi period. Now let's say we're playing in Japanese byoyomi, e.g., with 3 periods of each 12 seconds. It's our turn, we think for 10 seconds and then make a move. In this case I would have expected the game record to show that we thought about it for 10s (2s left) using the BL/WL property. I.e. I would have expected BL[2]OB[3] However, this is not what's happening when I look at some sgf files I got from KGS. What happens there is that after the move is made the clock resets to 12 seconds (the time available for the next move), and that value is stored in the sgf. So, once in Japanese byoyomi the sgf just shows a long list of BL[12]OB[3], which in my opinion is not very informative (all it tells us is that the move was completed within the period, but not exactly how much time it took). I'm tempted to think that this is a bug, but since it might also be a difference of interpretation of the spec I'd like to hear what others believe is the correct implementation. Best, Erik ___ 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] Japanese byoyomi in SGF files (possible KGS Bug?)
Hi All, I was just looking at how Japanese byoyomi time information is stored in sgf files. The spec at http://www.red-bean.com/sgf/properties.html tells us: Property: BL (WL) Function: Time left for black (white), after the move was made. Value is given in seconds. Property: OB (OW) Function: Number of black (white) moves left (after the move of this node was played) to play in this byo-yomi period. Now let's say we're playing in Japanese byoyomi, e.g., with 3 periods of each 12 seconds. It's our turn, we think for 10 seconds and then make a move. In this case I would have expected the game record to show that we thought about it for 10s (2s left) using the BL/WL property. I.e. I would have expected BL[2]OB[3] However, this is not what's happening when I look at some sgf files I got from KGS. What happens there is that after the move is made the clock resets to 12 seconds (the time available for the next move), and that value is stored in the sgf. So, once in Japanese byoyomi the sgf just shows a long list of BL[12]OB[3], which in my opinion is not very informative (all it tells us is that the move was completed within the period, but not exactly how much time it took). I'm tempted to think that this is a bug, but since it might also be a difference of interpretation of the spec I'd like to hear what others believe is the correct implementation. Best, Erik ___ 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
Are you looking at the initial state or the state after W j8 (after which B J7 is self-atari, and W lives)? Perhaps just show the position here to make sure we talk about the same one. On Tue, Jun 25, 2013 at 4:01 PM, ds d...@physik.de wrote: Hi, I really put all my reading skills into case1.sgf of two_safe_groups. the test file says black to move, and I really can not get an answer for w to win against B J7 (and oakfoam does not too:). (by the way, the sgf says w to move) Could anybody please give me a hint Detlef Am Mittwoch, den 20.03.2013, 12:39 + schrieb Aja Huang: Dear all, If you are interested, you can download the newest version of our regression test set (seki and two-safe-groups) at http://webdocs.cs.ualberta.ca/~shihchie/seki-and-two-safe-groups-regression-test.zip or in Fuego svn http://fuego.svn.sourceforge.net/viewvc/fuego/trunk/regression/name which contains the results of all participating programs including Crazy Stone, Zen, Steenvreter, pachi, ManyFaces, Gomorra and Fuego, etc. Kind regards, 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 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Computer Go at the European Go Congress
I am considering it as well, but I fear there are too many competing events in that period... Erik On Thu, May 2, 2013 at 6:58 PM, Petr Baudis pa...@ucw.cz wrote: Hi! On Thu, May 02, 2013 at 11:53:57AM +0100, Nick Wedd wrote: I would like to hear from the owners of strong programs whether they are planning to enter. If no strong program will be competing, I will not want to travel to Poland for the event. If there will be at least two strong programs competing, I will travel to Poland, and will offer to operate competing programs for those who cannot be there themselves, if this is to be permitted. I am considering to travel to Poland, but I am not decided yet whether I will go. I'm waiting for some sort of official announcement of the tournament with more detailed information; from http://egc2013.go.art.pl/page/go-tournaments I've understood that we could install Linux on the competition computer. -- Petr Pasky Baudis For every complex problem there is an answer that is clear, simple, and wrong. -- H. L. Mencken ___ 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] Supporting Japanese rules
On Fri, Feb 22, 2013 at 5:18 AM, Martin Mueller mmuel...@ualberta.ca wrote: ... Do people consider this a solved problem? I do. BTW I think it would be nice to have some kgs tournaments with Japanese rules; good for testing, and it might even make 9x9 a bit more interesting... Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Hardware advice
Personally I'd go for the 6344, or perhaps the 6238, but in general I agree; comparable Intel systems are ridiculously over-priced. Erik On Fri, Jan 11, 2013 at 9:07 PM, Rémi Coulom remi.cou...@free.fr wrote: Hi, We are considering investing into some high-end tournament hardware for Crazy Stone. It has to be a single machine. Right now, I am considering a server such as this one: http://www.thinkmate.com/System/RAX_QS5-4410 With 4 x Sixteen-Core AMD Opteron™ Model 6386 SE (2.8 GHz) It seems that Intel alternatives with a similar price are much less powerful. Is there anything I am missing? I checked that the clock efficiency of our Intel server is 1.4 better than our AMD bulldozer. I expect that hyperthreading does not help very much because of scalability problems, but I am not so sure. Any recommendation would be welcome. Thanks, 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
Re: [Computer-go] Board Sizes
Well, I still think mirroring is a bit easier on even size boards (you can't just play tengen to break it). But sure, even size boards are playable. Perhaps it's simplicity? In terms of complexity per intersection odd-size square boards are simplest, followed by even size square boards, and then the rectangular boards. 1xn Go is hell :-) Erik On Mon, Jan 7, 2013 at 3:23 PM, Don Dailey dailey@gmail.com wrote: So really then even size boards are perfectly playable.I would assume that odd sized boards have some characteristic that makes them more desirable, but it has nothing to do with mirror go. Don On Mon, Jan 7, 2013 at 9:02 AM, Vlad Dumitrescu vladd...@gmail.com wrote: Hi! On Mon, Jan 7, 2013 at 2:51 PM, Don Dailey dailey@gmail.com wrote: 2013/1/7 Don Dailey 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 ___ 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] Zen resignation positions
It doesn't look that easy to me. Have you tried playing it against Erica? Erik On Sun, Dec 2, 2012 at 3:02 PM, Aja Huang ajahu...@gmail.com wrote: In game 3 (Zen as W vs. So 8p), I don't understand why Zen didn't simply extend at G8 (move 24). That would be an easy win if Zen lived a group at that corner. Aja 2012/12/2 Hiroshi Yamashita y...@bd.mbn.or.jp Hi, Zen lost six games against pros in 9x9 on November 25. Each three pros played two games, black and white. I tried these six resignation positions by Aya 100k playouts. Game, Winrate for Aya 10.45 34th move, W to play, vs ICHIRIKI Ryo2p 20.33 53rd move, B to play, vs OHASHI Hirofumi 5p 30.14 34th move, W to play, vs SO Yokoku 8p 40.37 27th move, B to play, vs ICHIRIKI Ryo2p 50.56 34th move, W to play, vs OHASHI Hirofumi 5p 60.27 51st move, B to play, vs SO Yokoku 8p For Aya, game 1 and 5 are difficult. Maybe we need to understand these losses with smaller playouts? Official page(in Japanese) Computer Go challenges pros in 9x9 part 2, fighting without gloves. http://entcog.c.ooco.jp/entcog/event/event20121125.html sgf (player's info updated) http://entcog.c.ooco.jp/entcog/event/kifu_20121125.zip Newspaper report. (in Japanese) Pros won all games against computer in 9x9. http://www.asahi.com/igo/topics/TKY201211270576.html Following is abstract. Computers strength is close to pros in 9x9. But pros made position difficult intentionally, computer made mistakes and lost. Zen almost won its first game, but W 18th, 20th were deep reading moves, and saved the game. From this game result, pros guessed If we make one difficult space, computer might be rough on the other side.. So Ohashi 5 pro made a few ko-ish shapes in his first game, and he didn't start ko soon. Ohashi played two 9x9 games against Zen this March. It was 1 win 1 loss. He said I have studied well and played thinking of Zen's behavior. (He has written artilces about 9x9 Go in Japanese Go Journal.) Another report in Chinese, I can't read though. http://news.sina.com.tw/article/20121125/8393009.html Regards, Hiroshi Yamashita - Original Message - From: Hideki Kato hideki_ka...@ybb.ne.jp To: computer-go@dvandva.org Sent: Monday, November 26, 2012 7:31 AM Subject: Re: [Computer-go] Zen beat Ha,Yongil 5p with 4 stones (Re: TAAI details?) Thank you Hiroshi. The names of the players are swapped, sorry all. The games were: 1) Ichiriki 2p (B) vs Zen, 2) Ohashi 5p (W) vs Zen, 3) So 8p (B) vs Zen, 4) Ichiriki 2p (W) vs Zen, 5) Ohashi 5p (B) vs Zen, 6) So 8p (W) vs Zen. The time setting was: 20 min main time and 30 second for every move (no extra byo-yomi). Chinese rules and 7.0 komi were used. Zen's hardware was 4 pc cluster (12, 6, 6 and 6 cores) at 4 GHz, same as TAAI and Zen19D/S on KGS. Professinals used much longer time than Zen; more than 20 min in four games of six while Zen used 10 or 15 min for all games. Sometime they thought more than 5 min (upto 10) for a move, move 11 of the first game for example. On the first game, Zen would play J4 (hane) next at move 16 (cut) but actually played G3 and went into Sekito-shibori (two stone edge squeeze) course. Either showed about 80% winrate but Ichiriki 2p said J4 could lead a safer win for W. Actually Zen had sure chances to win; hard luck. Zen had chances to draw in some games in Black but selected much risky (actually losing) moves. I guess this caused by implementing draws by adding a third value, 0.5, to UCB. To play a draw move, all better looking moves (by prior) have to be refused, or proved worse than 0.5. This could take pretty long time in some positions. #Faking komi to 6.5 might help but a better solution possible? Hideki Hiroshi Yamashita: 313F02D8CF2F41D6AD89D1A42486DD59@x60: Hi, The game records can be downloaded from http://entcog.c.ooco.jp/entcog/event/kifu_20121125.lzh . I can not read that format. Can someone please translate it into I changed it to zip. http://www.yss-aya.com/kifu20121125.zip I was impressed So Yokoku 8p comment, Zen is apt to make a mistake when we make two places that may become ko. It makes game complex. Actual playing ko makes game simple. Didn't it win the first game? Sgf's player color is wrong. Black is Ichiriki 2p in first game. Zen lost all 6 games. Regards, 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] Congratulations to AyaMC!
Hi Nick, In round 2 Steenvreter's kgsGtp failed to connect to the server due to a problem with dns lookup. When I saw what had happened I managed to start another instance on an old machine which I hadn't used in a long time (running some rather old versions of Steenvreter and kgsGtp). I was hoping it would play the entire game against gnugo in byo yomi. If it hadn't been for some additional netlag it even might have won (the winning percentage was still ok at the end). In round 3 the crippled version of Steenvreter started to play Aya (not Many Faces). Somewhere during that game the original machine came back online. Both seemed to be kicking out the other repetitively, so I killed the slow one, but then it was already too late. Best, Erik On Mon, Nov 5, 2012 at 1:15 PM, Nick Wedd n...@maproom.co.uk wrote: Congratulations to AyaMC, winner of yesterday's 19x19 KGS bot tournament! The results are at http://www.weddslist.com/kgs/past/87/index.html Please let me know of any errors or 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] double ko seki on KGS
On Sun, Oct 21, 2012 at 11:38 AM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: Hi, I saw an interesting game. http://files.gokgs.com/games/2012/10/21/globax-AyaMC2.sgf In 337th move, W passed, but B took ko in double ko seki. And in 339th move, W had two forbidden ko points by super-ko rule. And B won finally. Is double ko seki not available on super-ko rule? Of course double ko seki is available; White's pass here was simply a blunder due to the positional superko rule. Erik ___ 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 Tue, Oct 2, 2012 at 11:40 AM, Rémi Coulom remi.cou...@free.fr wrote: 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 The 9x9 tournament has integer komi; that line is the same for all sizes. BTW I think the KGS superko rule shouldn't be a problem in this type of tournament (round-robin). When such a situation occurs the game simply ends due to a repetition / 'illegal' move. Since play doesn't have to continue the tournament organizers can just assign the correct result. Any info on registration fee's, does one have to attend the congress, etc.? Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Zen!
I tend to agree. Nick, in my opinion, unless there are urgent reasons like delaying a fixed schedule, the choice to resign should be made by the program. Any kibitzer concluding that the position is hopeless is free to move on to watch another game, and the program that is certain of its win does not have to keep calculating; it can simply reduce it's load to the bare minimum (thus reducing your feared contribution to global warming). Further, if you go this path, where does it end? Will you stop weak programs from entering? Maybe it would be more interesting to run a McMahon tournament instead, to avoid such uninteresting games between totally unbalanced opponents? Erik On Sat, Sep 15, 2012 at 6:06 PM, steve uurtamo uurt...@gmail.com wrote: the business about ending games in completely hopeless positions -- i'm not sure that makes the most sense. i realize how aggravating it can be for observers, especially in such a long game, but i'm not sure that it's the right decision. s. On Sat, Sep 15, 2012 at 8:53 AM, Nick Wedd n...@maproom.co.uk wrote: Congratulations to Zen19S, winner of the Slow KGS bot tournament which ended ten days ago! My report is at http://www.weddslist.com/kgs/**past/S12.2/index.htmlhttp://www.weddslist.com/kgs/past/S12.2/index.htmland I look forward, as usual, to receiving your comments and reports of errors. I apologise for the long delay in uploading this report. Nick -- Nick Wedd n...@maproom.co.uk __**_ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/**mailman/listinfo/computer-gohttp://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] A question on seki scoring in Japanese rules
The simple answer is Robert's recommendation to capture before ending the game. Once you get to the scoring phase every stone in the seki should be assumed alive (even if, e.g., it has only one liberty). It is up to the players to decide which pieces are dead (and therefore can be captured) before the game ends. This is not a trivial decision because it may destroy the seki (i.e., you could loose a semeai). For fun look up Hanezeki Erik On Sun, Sep 9, 2012 at 4:30 PM, fastgo fas...@gmail.com wrote: Thanks. I thought it is a common situation, there would be simple answers, like in the case of seki eye with one dead piece, it is 0,1,2 or 3 etc. On Sep 9, 2012 10:20 AM, Robert Jasiek jas...@snafu.de wrote: On 09.09.2012 16:09, fastgo wrote: Japanese rules. In a case of seki where one side has one or two surrounded dead stones, how are those scored? It depends on which Japanese ruleset is used and on the procedural moment when the question is being studied for a position. It can depend on referee arbitration. http://home.snafu.de/jasiek/j_**verbal_status.pdfhttp://home.snafu.de/jasiek/j_verbal_status.pdf http://home.snafu.de/jasiek/**j1989c.htmlhttp://home.snafu.de/jasiek/j1989c.html http://home.snafu.de/jasiek/**wagc.htmlhttp://home.snafu.de/jasiek/wagc.html http://home.snafu.de/jasiek/**wagci.htmlhttp://home.snafu.de/jasiek/wagci.html http://home.snafu.de/jasiek/**wagcmod.htmlhttp://home.snafu.de/jasiek/wagcmod.html http://home.snafu.de/jasiek/**wagcinf.htmlhttp://home.snafu.de/jasiek/wagcinf.html http://home.snafu.de/jasiek/**wagcflaw.htmlhttp://home.snafu.de/jasiek/wagcflaw.html In short, capture (possibly throw-in and capture again) before the first succession of passes is my recommendation. -- robert jasiek __**_ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/**mailman/listinfo/computer-gohttp://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] A question on seki scoring in Japanese rules
On Sun, Sep 9, 2012 at 5:16 PM, Robert Jasiek jas...@snafu.de wrote: On 09.09.2012 16:56, Erik van der Werf wrote: Once you get to the scoring phase every stone in the seki should be assumed alive (even if, e.g., it has only one liberty). It is up to the players to decide which pieces are dead This answer does not hold for Japanese rules in general. What not exactly? The first line or the (partially quoted) second line? Anyway, it is meant as advice for programming a Go bot. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] A question on seki scoring in Japanese rules
On Sun, Sep 9, 2012 at 6:30 PM, Robert Jasiek jas...@snafu.de wrote: On 09.09.2012 17:29, Erik van der Werf wrote: Once you get to the scoring phase every stone in the seki should be assumed alive (even if, e.g., it has only one liberty). It is up to the players to decide which pieces are dead This answer does not hold for Japanese rules in general. The first line or the (partially quoted) second line? Both sentences. J1989, double ko seki, the ko stones are dead. J1989 or WAGC: the players do not have free choice to decide but they must determine status according to the rules. Thx. While I find the concept 'dead in seki' a bit absurd, it's interesting to know it exists in some variations of Japanese rules. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] A question on seki scoring in Japanese rules
On Sun, Sep 9, 2012 at 6:30 PM, Robert Jasiek jas...@snafu.de wrote: J1989, double ko seki, the ko stones are dead. Looking into this a bit more, it seems that although the stones are officially termed 'dead', for counting points they are not removed from the board because they are not in territory. Is that correct? Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] An unusual seki
On Thu, Aug 16, 2012 at 1:06 AM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: I understood why I and bots were banned for a while. Oh, interesting, did this lead to a kgs ban? Why exactly was that? Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] An unusual seki
I saw Nick's email, but booting is not the same as banning. An actual ban is an implicit acknowledgement of a flaw in the kgs scoring protocol. Just thought it might be interesting to hear some details (especially if and how it was fixed). Anyway, all this seems rather strange. Bots only need one resumption. After that all remaining stones are assumed to be alive by rule, so there's no need to query the bot again for anything. The game can simply end without any need for an admin to act. One could make this a bit more flexible by increasing the max number of resumptions, but the principle remains the same; there really shouldn't be a need for an admin to get involved to get a game to end. Erik On Sun, Aug 19, 2012 at 6:26 PM, Michael Williams michaelwilliam...@gmail.com wrote: See original email from Nick: Yesterday, a KGS game between Blubbel 3d and AyaBot4 2k, SGF file below, ended with an unusual kind of seki. AyaBot4 marked its opponent's stones in the seki as dead, and was eventually booted by an admin for mis-marking stones (as a way of getting the game to end). As all eleven AyaBots use the same IP address, they all got booted - and an hour later, all simultaneously tried to log in again. On Sun, Aug 19, 2012 at 8:45 AM, Erik van der Werf erikvanderw...@gmail.com wrote: On Thu, Aug 16, 2012 at 1:06 AM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: I understood why I and bots were banned for a while. Oh, interesting, did this lead to a kgs ban? Why exactly was that? Erik ___ 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] Congratulations to Zen!
Indeed Steenvreter wasn't running on Pachi's hardware. As in other recent tournaments I used an Opteron system with 46 threads at 2.2 Ghz. It's not as powerful as Zen's minicluster, but at least it gets a bit closer than using that (single cpu) i7 machine. Regarding the komi; to me it has been quite clear for a long time that 7.5 is too high for the 9x9 board. In my mind the issue was already settled in 2008 (at least from what I remember talking about it with David and some other guys in Beijing), but somehow it still took a surprisingly long time to settle in. (BTW I suspect 7.5 komi to be too high for all larger boards, because I see no obvious reason why komi would start increasing again with board size). Anyway, 7.0 might be fair, but if it's still too high then 6.5 and 5.5 probably won't make much difference under Chinese rules. Is anyone (besides me) willing to take the next step in Computer Go evolution: Japanese style rules, territory scoring + 6.5 komi. Because of the nearly-doubled resolution in likely possible scores it might keep 9x9 Go interesting for some more time to come (and perhaps even give Nick a chance again to make some comments on the games ;-)). Best, Erik On Mon, Aug 13, 2012 at 12:25 PM, ds d...@physik.de wrote: Thanks a lot!! great tournament. stv was probably not running pachi (as you wrote in the hardware section:) As far as I know he runs on i7-3930.. Detlef Am Montag, den 13.08.2012, 11:11 +0100 schrieb Nick Wedd: Congratulations to Zen, winner of yesterday's bot tournament! My (very short) report is at http://www.weddslist.com/kgs/past/85/index.html As usual, I hope you will send me your corrections and comments. (One of you has already suggested that I use points instead of wins as a column header in the cross-table, because a jigo counts as half a point. But I prefer wins - I consider that a jigo is half a win, and there are many things (such as SOS) which could be described as points.) Nick ___ 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] Kaś Cup - results and prizes
On Thu, Aug 9, 2012 at 11:14 AM, Petr Baudis pa...@ucw.cz wrote: On Wed, Aug 08, 2012 at 09:08:47PM +0200, ds wrote: Hyperthreading does the trick, I have the experience it increases the performance by about 10%. I think this is due to waiting for RAM I/O or things like that Yes. With hyperthreading, performance per thread goes down significantly, but total performance goes up by about 15%. In the Pentium 4 era, hyperthreading did not usually pay off, but with i7, its performance is much better. The basic idea is that there are two instruction pipelines that share the same ALU and other processor units; if one of the pipelines stalls (usually due to memory fetch), the other can use the ALU in the meantime, or the two threads may use different parts of the CPU altogether based on what the instructions do. 10-15%, really, that low? For my program (on an i7-3930K, going from 6 to 12 threads) it is more in the order of 40% extra simulations per second. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Kaś Cup - results and prizes
I don't have an i7-2600, but I could run oakfoam on the 3930. I just downloaded it and it does compile. If you give me a list of gtp commands to run the benchmark, then I will send you the output back. Erik On Thu, Aug 9, 2012 at 12:38 PM, ds d...@physik.de wrote: This is very interesting, I have not more than 10% with oakfoam on i7-2600K. Would be interesting if it is the processor or if you e.g. access more often memory instead of cache due to your code... Do you have the chance to run your program on a i7-2600? or do you have to much time and try https://bitbucket.org/francoisvn/oakfoam/wiki/Home on your i7-3930. If so, I would be very much interested in the number you get in the beginning of a 19x19 game without book:) Detlef Am Donnerstag, den 09.08.2012, 12:16 +0200 schrieb Erik van der Werf: On Thu, Aug 9, 2012 at 11:14 AM, Petr Baudis pa...@ucw.cz wrote: On Wed, Aug 08, 2012 at 09:08:47PM +0200, ds wrote: Hyperthreading does the trick, I have the experience it increases the performance by about 10%. I think this is due to waiting for RAM I/O or things like that Yes. With hyperthreading, performance per thread goes down significantly, but total performance goes up by about 15%. In the Pentium 4 era, hyperthreading did not usually pay off, but with i7, its performance is much better. The basic idea is that there are two instruction pipelines that share the same ALU and other processor units; if one of the pipelines stalls (usually due to memory fetch), the other can use the ALU in the meantime, or the two threads may use different parts of the CPU altogether based on what the instructions do. 10-15%, really, that low? For my program (on an i7-3930K, going from 6 to 12 threads) it is more in the order of 40% extra simulations per second. Erik ___ 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] Seki in playouts
On Tue, Aug 7, 2012 at 10:34 PM, Michael Williams michaelwilliam...@gmail.com wrote: Could anyone describe the basic implementation for detecting some seki in a fast enough way to be used in playouts? I understand that it's not practical to detect them all. Just prevent self-atari unless it makes a basic nakade shape and the surrounding group doesn't have eye potential elsewhere. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Kas Cup
Lukasz' current rule gives a big advantage to i7 (low core count, 2 threads per core, high clock rate, all the opposite for Opteron systems). In my experience performance per thread per GHz for a fully loaded system is quite similar for a wide range of processor types (I tested this for Opteron, i7 and an older core2). Erik On Fri, Jul 27, 2012 at 7:31 AM, David Fotland fotl...@smart-games.comwrote: 2 threads is not worth nearly as much as two cores. ** ** David ** ** *From:* computer-go-boun...@dvandva.org [mailto: computer-go-boun...@dvandva.org] *On Behalf Of *Erik van der Werf *Sent:* Thursday, July 26, 2012 2:09 PM *To:* computer-go@dvandva.org *Subject:* Re: [Computer-go] Kas Cup ** ** On Wed, Jul 11, 2012 at 6:43 PM, Łukasz Lew lukasz@gmail.com wrote:* *** I'm assuming that hyperthreading is OK, so a core i7 would count as 4 cores, but use 8 threads. Yes. ** ** So what about AMD cores? What about overclocking? ** ** I think limiting on nr_of_threads X clock_rate would be more fair. ** ** Erik ** ** ___ 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] Kas Cup
Ok, so what if the closest I can get is a 6-core i7. How many threads may I start on that one? Erik On Fri, Jul 27, 2012 at 6:48 PM, Łukasz Lew lukasz@gmail.com wrote: On Fri, Jul 27, 2012 at 2:15 AM, Erik van der Werf erikvanderw...@gmail.com wrote: Lukasz' current rule gives a big advantage to i7 (low core count, 2 threads per core, high clock rate, all the opposite for Opteron systems). In my experience performance per thread per GHz for a fully loaded system is quite similar for a wide range of processor types (I tested this for Opteron, i7 and an older core2). These rules are a compromise between simplicity and the goal, which is to measure algorithms instead of hardware. Most important effect of limiting of number of cores is to cut a huge advantage of using big clusters. I am less concerned about relatively minor differences of architectures or GHz. Erik On Fri, Jul 27, 2012 at 7:31 AM, David Fotland fotl...@smart-games.com wrote: 2 threads is not worth nearly as much as two cores. David From: computer-go-boun...@dvandva.org [mailto:computer-go-boun...@dvandva.org] On Behalf Of Erik van der Werf Sent: Thursday, July 26, 2012 2:09 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Kas Cup On Wed, Jul 11, 2012 at 6:43 PM, Łukasz Lew lukasz@gmail.com wrote: I'm assuming that hyperthreading is OK, so a core i7 would count as 4 cores, but use 8 threads. Yes. So what about AMD cores? What about overclocking? I think limiting on nr_of_threads X clock_rate would be more fair. Erik ___ 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 -- Łukasz ___ 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] Kas Cup
On Wed, Jul 11, 2012 at 6:43 PM, Łukasz Lew lukasz@gmail.com wrote: I'm assuming that hyperthreading is OK, so a core i7 would count as 4 cores, but use 8 threads. Yes. So what about AMD cores? What about overclocking? I think limiting on nr_of_threads X clock_rate would be more fair. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] test case on divide conquer
On Wed, Jul 11, 2012 at 5:03 PM, terry mcintyre terrymcint...@yahoo.com wrote: Am I misreading this? yes, neither side can approach. Capturing in the big eyes give too many liberties. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Fw: test case on divide conquer
On Wed, Jul 11, 2012 at 9:52 PM, terry mcintyre terrymcint...@yahoo.com wrote: From: Erik van der Werf erikvanderw...@gmail.com On Wed, Jul 11, 2012 at 5:03 PM, terry mcintyre terrymcint...@yahoo.com wrote: Am I misreading this? yes, neither side can approach. Capturing in the big eyes give too many liberties. Let's start with w e9 - b is in atari, but w has liberties at c7, g9, and e5 - if b plays any of these, a b group is captured. b d11 captures the w rabbity six; w throws in at c12. b cannot make two eyes. correct so far... Nor can b play the other 3 liberties. wrong, White just has to make an approach move for each remaining shared liberty (i.e. capture a nakade) whether b plays inside the upper right corner or passes, w kills the upper right corner. No, White cannot kill any of the Black's groups. Trust me, even my program can solve this :-) White's liberty count after your first 3 moves is : 2 shared liberties +2 approach moves + 6-point eye filled with 5 stones = 11 liberties Black's weakest group has the same number of liberties, but Black has sente so he wins. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Fw: test case on divide conquer
small correction On Wed, Jul 11, 2012 at 10:59 PM, Erik van der Werf erikvanderw...@gmail.com wrote: ... Nor can b play the other 3 liberties. wrong, White just has to make an approach move for each remaining shared liberty (i.e. capture a nakade) should of course be: Black just has to make an approach move for each remaining shared liberty (i.e. capture a nakade) ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] TCGA 19x19 Computer Go Tournament, error report
Hi Jirong, Wow the 19x19 results now look really bad for Steenvreter. I know it had a poor start, but I thought it could easily recover because there were many rounds left. How did so many of those timed out and forfeited games end up counting against my program? I don't mind if you just cancel the 19x19 tournament. If you reschedule, please consider using faster time settings; this was way too slow IMO. Best, Erik On Sun, Jul 1, 2012 at 4:29 AM, Jirong jirong...@gmail.com wrote: Dear David, We will reschedule it again or finish the remaining games on another day. regards, Jirong On Sun, Jul 1, 2012 at 9:29 AM, David Fotland fotl...@smart-games.com wrote: Perhaps the best thing to do is to cancel the 19x19 and schedule it again later? regards, David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of David Fotland Sent: Saturday, June 30, 2012 6:15 PM To: n...@maproom.co.uk; computer-go@dvandva.org Subject: Re: [Computer-go] TCGA 19x19 Computer Go Tournament, error report It looks like in rounds 10-15 every match timed out, which put the tournament back on schedule. These matches should not be counted in the official totals. David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Nick Wedd Sent: Saturday, June 30, 2012 4:19 PM To: computer-go@dvandva.org Subject: [Computer-go] TCGA 19x19 Computer Go Tournament, error report This report is about incidents in the ongoing TCGA 19x19 Computer Go Tournament, now running on KGS. I am writing this for the record, so that the players and organisers will be aware of all the information that I have. These statements are made for the record. I hope I can be regarded as impartial. I am only making statements of fact anyway. __Incident 1. During round 2, I removed 'ntnu' from the tournament, at the request of its operator. I hoped that this would reduce the length of the tournament from 36 rounds to a more manageable 30, which would allow it to end before the start of the July KGS tournament. However, it seems that that is not how the scheduler for a multiple- round-robin works. Round 10, with eight players, has three games and two byes, rather than the four games that one might expect. I assume that the entire schedule was drawn up in advance, and one of the two byes was scheduled as a game against the now-absent 'ntnu'. I do not know how to avoid a clash with the schedule of the KGS July bot tournament. I am now too tired to think sensibly about it, and will go to bed soon. I will leave 'guxxan' to think about it. __Incident 2. During round 3, 'ajahuang' saw that Fuego19, which he was operating, had a hopeless position against nomitan440, and proposed logging in and resigning for it. If I had been more alert, I would have realised what was likely to happen, and told him not to; but I said nothing. He logged in to Fuego19's account, and resigned for it. The KGS tournament system observed a human using a bot's account in a bot tournament, regarded this as an attempt to cheat, and marked the game as lost by Forfeit. Thus this round 2 game has two different results: KGS sees it as lost by Fuego19 by resignation, while the KGS tournament system sees it as lost by Fuego19 by Forfeit. There is no significant problem so far, this outcome was what I had expected. However, the tournament was running behind the optimistic 25-minutes-a- round specified when it was set up. So when this round 2 game ended (all the other round 2 games had already ended), round 3 began immediately. And ajahuang was still logged in to Fuego19's account. So Fuego19's round 3 game (against AyaMC5) was also treated, by the tournament system, as lost by Fuego19, by Forfeit. But the game continued to be played. When the other round 3 games were finished, round 4 started (the tournament system regarding the Fuego19/AyaMC5 game as already decided). I realised that Fuego19 and AyaMC we still playing in their round 3 game and had not joined their round 4 games, so, to free them up and let them play in round 4, I used my admin power to kill this game. I believe that Fuego19 had been winning it at the time. I killed it rather too slowly, so that when AyaMC5 joined its round 4 game against Zen19S, it has already lost one of its 15-second overtime periods. I apologise to AyaMC5's operator for this. Fuego19 did not suffer loss of an overtime period, as it had a bye in round 4. 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
Re: [Computer-go] oakfoam/Zen9 game
Why guess? Yamato, don't you or Hideki keep log files with the output of kgsgtp? There should be some lines like: INFO: Disagreement over tournament scoring. Switching to cleanup mode. ... FINEST: Command sent to engine: undo ... FINEST: Command queued for sending to engine: kgs-genmove_cleanup b ... Erik On Mon, Feb 13, 2012 at 10:53 AM, Nick Wedd n...@maproom.co.uk wrote: On 13/02/2012 00:10, Yamato wrote: I guess the server did not send the clean-up command to Zen. This seems unlikely. The server usually sends the clean-up command to both players. My guess is: the server sent the clean-up command to both players oakfoam passed the server took this pass, with the previous pass that was part of the game, to be two consecutive passes, and therefore the end of the clean-up phase. I have written to wms about this. 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] oakfoam/Zen9 game
The cleanup variation doesn't show a (extra) black pass, so that's indeed suspicious. But maybe kgs just failed to record the third consecutive pass, or for some reason did not create a new node when black accidentally passed again (after undoing twice). It is not clear to me how many consecutive passes are supposed to be undone when the game goes into cleanup mode. Looking at a few sgf files where my program got into cleanup mode (btw always against mgf) I can actually see some different behaviors... Maybe kgs tries to be smart in determining who has to do the capture? (and then perhaps something gets confused on the game-end criterion) I think to be sure one would have to look at Zen's log file containing the communications by kgsgtp. Erik On Sun, Feb 12, 2012 at 2:23 PM, Nick Wedd n...@maproom.co.uk wrote: In my report http://www.weddslist.com/kgs/past/80/index.html on the last KGS bot tournament, I wrote In round 13, Zen9 (as Black) and then oakfoam passed ... . Zen has won: the 14-stone white group ... is dead ... . However they disagreed about the status of the 14-stone group, and the game entered the clean-up phase. Oakfoam again passed, and the game was immediately declared over, with all the stones on the board counted as alive. The game was therefore counted as a win for oakfoam. It seems to me that Black was never given a chance to show that it could capture the dead group. I shall check, and if my understanding is correct, I shall report this to KGS programmer 'wms' as a defect in the clean-up mechanism. I would be grateful if someone could check my interpretation of what happened. Is this just an example of one of the bots handling the clean-up wrong? Or is it the server itself that got it wrong? If it's the server, I must report it to wms. 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 SteenVreter!
Yes, Steenvreter used a bigger machine (from the Maastricht games group). I used the exact same setup as in the december tournament (which it also won) and in the computer olympiad. Sunday I also tested another machine. The machine in Maastricht (Opteron system, running 46 threads @ 2.2 GHz) is around 2.5 times faster (in terms of simulations per second) than a single i7-3930K @ 3.2 GHz (12 threads). So, performance per GHz between intel and amd appears to be almost identical... Erik On Tue, Feb 7, 2012 at 4:25 PM, Aja Huang ajahu...@gmail.com wrote: Congratulations to SteenVreter! Nick, thanks for the report. SteenVreter is strong. It won all the games except the 3 losses against Zen. By the way, Erik said SteenVreter was running on a big hardware (24-core SMP?), not a 4-core machine. Aja -原始郵件- From: Nick Wedd Sent: Tuesday, February 07, 2012 7:57 AM To: computer-go@dvandva.org Subject: [Computer-go] Congratulations to SteenVreter! Congratulations to SteenVreter (stv), winner of Sunday's KGS bot tournament, with 20 wins, one jigo, and three losses from 24 games! My report is at http://www.weddslist.com/kgs/past/80/index.html I hope you will email me about any mistakes in it, also with any 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] Congratulations to SteenVreter!
On Tue, Feb 7, 2012 at 5:50 PM, David Fotland fotl...@smart-games.com wrote: Thanks Nick. Interesting that ManyFaces beat Zen 2-1, beat Pachi 2-1, but lost all four games to Aya. ManyFaces still doesn't understand ties in the playouts, so it gave up one Jigo. Also interesting that Zen beat Stv head to head, but Stv was far more consistent against the weaker programs. Zen must still have some bugs. Well it was 3.5 - 1.5, and 3 games had a rather similar opening. Not really enough for strong conclusions. Anyway, I'm happy ManyFaces and Pachi did well against Zen :-) Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
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
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
On Tue, Jan 3, 2012 at 2:27 PM, Robert Jasiek jas...@snafu.de wrote: On 03.01.2012 13:58, Ingo Althöfer wrote: 10 games against strong bots within 30 days would be one possible condition; How? The clicking fastest to accept a game match is still the problem. So the problem is that too many humans want to play strong bots? Perhaps it would help if kgs provided an option in the kgsgtp config to limit the rank difference where challenges are accepted. Humans can be picky about their opponents; maybe bots should also be allowed into that game :-) Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] level of MCTS programs depending on the number of sims in 9x9 (unlimited time setting for humans)
Let me guess: 1000 simulations per move ~= 1 Dan ? Erik On Sat, Dec 10, 2011 at 2:19 PM, Olivier Teytaud teyt...@lri.fr wrote: Hi Brian; thanks for sharing these results. I'll try to post here results in Kyu / Dan. For the first results I've seen it's a bit surprising to me; in 9x9 very small numbers of simulations have unexpectedly good results... but I'll wait for more stats for sharing as it's really surprising to me :-) (experiences with humans in the loop take so much time...) Incidentally, if you want to contribute, here a version of MoGo which adapts itself to your level; from the tables of results it generates you can see your level in MC simulations: http://teytaud.over-blog.com/article-a-linux-program-for-evaluating-your-level-in-go-91946330.html (sorry, only for Linux...) Best regards, Olivier 2011/11/19 Brian Sheppard sheppar...@aol.com I have been running Fuego 0.4.1 and Mogo3 on CGOS 9x9 for many thousands of games. Here is the data: Fuego0.4.1-100K,p 2657 Fuego0.4.1-30K,p 2520 Fuego0.4.1-10K,p 2376 Fuego0.4.1-3K,p 2205 Fuego0.4.1-1K,p 2024 Fuego0.4.1-300,p 1756 Mogo3MC90K,p 2428 Mogo3MC30K,p 2304 Mogo3MC10K,p 2169 Mogo3MC3K,p 2008 Mogo3MC1K,p 1823 From: computer-go-boun...@dvandva.org [mailto:computer-go-boun...@dvandva.org] On Behalf Of Olivier Teytaud Sent: Thursday, November 17, 2011 11:11 PM To: computer-go Subject: [Computer-go] level of MCTS programs depending on the number of sims in 9x9 (unlimited time setting for humans) Hi; Does anyone roughly know the level of a MCTS program in 9x9 depending on the number of simulations per move ? I know there are plenty of drawbacks to any such correspondence, that it depends on your style, on the time settings (I here assume the human has very long time...) and so on, but any rough estimate would be better than nothing :-) I would roughly say: 500 simulations = KGS 10 kyu 3 000 simulations = 5 kyu 15 000 simulations = KGS 1 Dan 3 000 000 simulations = pro level ...but I have no clear idea of this :-) (please don't spend too much time on the 1000 reasons for which such a correspondance does not make too much sense, we all know it :-) ) Best regards, Olivier ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- = Olivier Teytaud -- olivier.teyt...@inria.fr TAO, LRI, UMR 8623(CNRS - Universite Paris-Sud), bat 490 Universite Paris-Sud F-91405 Orsay Cedex France http://0z.fr/EJm0g (one of the 56.5 % of french who did not vote for Sarkozy in 2007) ___ 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
I don't like it! So not only did they raise the price from 25 to 40 euro *, but now they even want to charge triple for 3 inferior mini-tournaments! In my opinion a one day 13x13 tournament really shouldn't be priced the same as a full-scale 5-day event (e.g., such as in chess). I don't mind some diversity in board size, but it's all still the same game. With this pricing scheme I'd prefer to strike 9x9 and 13x13, and just spend 5 days on 19x19 Go. And btw what's up with those 'fresh' amateurs??? The distinction used to be quite clear: if you had no direct or indirect commercial interests then you were an amateur. I haven't participated in recent years, does that make me fresh again? or do I still stink because I went to Beijing? Perhaps some deodorant will do the trick... Erik * not counting ICGA membership On Thu, Sep 22, 2011 at 7:28 PM, Rémi Coulom remi.cou...@free.fr wrote: 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 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Rules for the Computer Olympiad
Hi Martin, Good question, I don't know. If Hideki doesn't know either (he compiled the last version based on my original) then it probably means this has not been decided yet. I suppose the default would be to use last year's rules. There is usually a players meeting just before the tournament starts, to formalize the details, but for significant changes that would at least for some be too late. Perhaps we should try to see if there is consensus about what we would like this year? My proposal would be to change the komi for 9x9 Go to 7.0, and leave the rest as it was last year. (So that would mean 19x19: 45m., 7.5 komi, 13x13: 30m., 7.5 komi, 9x9: 10m., 7.0 komi, remote play allowed.) What are your preferences? Best, Erik On Sat, Sep 17, 2011 at 1:00 AM, Martin Mueller mmuel...@ualberta.ca wrote: Does anyone here know which rules will be used for the Olympiad? Especially regarding: - komi - time limits - remote play thanks Martin ___ 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] Rules for the Computer Olympiad -- Remote Play
I think Go will always have its own set of game-specific rules for things like komi and timing (and I'd be happy enough just to see consensus there). However, regarding remote play I don't see a good reason why Go must deviate from the 'general' rules. Personally, I don't much like remote play either, but it has some interesting aspects. I guess the question is what do we want to identify: the strongest program or the strongest artificial playing entity? Erik On Sat, Sep 17, 2011 at 4:27 PM, Richard J. Lorentz lore...@csun.edu wrote: Very often the rules at the Olympiads tend to be uniform across all the games. This, of course, need not be the case but should that happen again this year I would like to speak against remote play. For the smaller games that I participate in, part of the fun and the challenge is seeing how my program on my little laptop fares against your program on your laptop. Especially now that MCTS is being used in so many programs, multiple and faster CPU's with bunches of memory really do give a significant advantage, and an unfair one to the many of us who do not have access to those kinds of resources. I would like to suggest that either we make it clear that go has its own set of rules (in which case this discussion needs to find an appropriate home) or we no longer allow remote play. -Richard On 9/17/2011 5:59 AM, Erik van der Werf wrote: Hi Martin, Good question, I don't know. If Hideki doesn't know either (he compiled the last version based on my original) then it probably means this has not been decided yet. I suppose the default would be to use last year's rules. There is usually a players meeting just before the tournament starts, to formalize the details, but for significant changes that would at least for some be too late. Perhaps we should try to see if there is consensus about what we would like this year? My proposal would be to change the komi for 9x9 Go to 7.0, and leave the rest as it was last year. (So that would mean 19x19: 45m., 7.5 komi, 13x13: 30m., 7.5 komi, 9x9: 10m., 7.0 komi, remote play allowed.) What are your preferences? Best, Erik On Sat, Sep 17, 2011 at 1:00 AM, Martin Muellermmuel...@ualberta.ca wrote: Does anyone here know which rules will be used for the Olympiad? Especially regarding: - komi - time limits - remote play thanks Martin ___ 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] RAVE uses all moves in playout?
Hi Hiroshi, I think the problem with adding moves of the opponent is that they don't come from the same distribution. E.g., imagine if Black is winning 90% then it's normal rave moves are also winning 90% on average (so mixing with the direct statistics doesn't cause a big change on average regardless of the mixing ratio). However, once you start to add White's moves the distribution is reversed (on average only winning 10%) and together those statistics will average roughly on 50%, which causes a systematic bias. The intersections where both sides want to play first will of course still stand out, so that's why reversing the sign helps at least a bit. However, to get this to work really well you need to find a clever way to get the bias down... Expand on this a bit and you have yourself a new rave paper ;-) Best, Erik On Fri, Aug 12, 2011 at 7:18 PM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: Hi Erik, I think the usual procedure is to update only moves by the same player. However, if you also want to use the opponent's moves (which is not an unreasonable idea), then you should probably reverse their sign (because those moves lost), so you only get something like: I tried this idea. 9x9 (1000games 6000po against Fuego 1po) 0.516 RAVE normal (same player only) 0.211 RAVE Black and White 0.264 RAVE Black and White, reverse their sign 0.092 RAVE not same player only, reverse their sign 19x19 (1000games 700po agaist GnuGo Lv0) 0.570 RAVE normal (same player only) 0.013 RAVE Black and White 0.078 RAVE Black and White, reverse their sign 0.141 RAVE not same player only, reverse their sign Normal is clearly better, but reverse their sign looks a bit meaningful. In a B node, update only B moves - and only if there's no W move at same location before B's move. I use this in 9x9, but I ignore all W moves in 19x19. I got a little better result from this. Regards, Hiroshi Yamashita - Original Message - From: Erik van der Werf erikvanderw...@gmail.com To: computer-go@dvandva.org Sent: Thursday, August 11, 2011 8:42 PM Subject: Re: [Computer-go] RAVE uses all moves in playout? On Thu, Aug 11, 2011 at 3:43 AM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: Hi, I have a question about RAVE. For example, there are 5 moves a,b,c,d,e. Black to play. In playout, Black plays a, White plays b, Black plays c, W plays d and B plays e. Then game is over, result is B win. B:a W:b B:c W:d B:e ... B win, result = +1 I update only B moves RAVE, like this. RAVEcount(a) += 1, RAVEwins(a) += 1 RAVEcount(c) += 1, RAVEwins(c) += 1 RAVEcount(e) += 1, RAVEwins(e) += 1 But on Sylvain's paper Figure4, It looks like updating all moves Black and White, like this. RAVEcount(a) += 1, RAVEwins(a) += 1 RAVEcount(c) += 1, RAVEwins(c) += 1 RAVEcount(e) += 1, RAVEwins(e) += 1 RAVEcount(b) += 1, RAVEwins(b) += 1 RAVEcount(d) += 1, RAVEwins(d) += 1 Is this right RAVE? I have misunderstood long time. Hi Hiroshi, I think the usual procedure is to update only moves by the same player. However, if you also want to use the opponent's moves (which is not an unreasonable idea), then you should probably reverse their sign (because those moves lost), so you only get something like: RAVEcount(b) += 1 RAVEcount(d) += 1 Best, Erik ___ 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] RAVE uses all moves in playout?
On Thu, Aug 11, 2011 at 3:43 AM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: Hi, I have a question about RAVE. For example, there are 5 moves a,b,c,d,e. Black to play. In playout, Black plays a, White plays b, Black plays c, W plays d and B plays e. Then game is over, result is B win. B:a W:b B:c W:d B:e ... B win, result = +1 I update only B moves RAVE, like this. RAVEcount(a) += 1, RAVEwins(a) += 1 RAVEcount(c) += 1, RAVEwins(c) += 1 RAVEcount(e) += 1, RAVEwins(e) += 1 But on Sylvain's paper Figure4, It looks like updating all moves Black and White, like this. RAVEcount(a) += 1, RAVEwins(a) += 1 RAVEcount(c) += 1, RAVEwins(c) += 1 RAVEcount(e) += 1, RAVEwins(e) += 1 RAVEcount(b) += 1, RAVEwins(b) += 1 RAVEcount(d) += 1, RAVEwins(d) += 1 Is this right RAVE? I have misunderstood long time. Hi Hiroshi, I think the usual procedure is to update only moves by the same player. However, if you also want to use the opponent's moves (which is not an unreasonable idea), then you should probably reverse their sign (because those moves lost), so you only get something like: RAVEcount(b) += 1 RAVEcount(d) += 1 Best, Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] testing improvements
On Thu, Aug 4, 2011 at 6:57 PM, Vlad Dumitrescu vladd...@gmail.com wrote: The scores towards gnugo are almost identical, but the two fuegos score 449-415, which is 52% and the 95% confidence is ~3%, i.e. ~10 ELO. That 3% is not a 95% confidence interval, more like 1 standard deviation... (so nothing with high confidence yet) Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] EGC computer-go tournament 9x9
On Mon, Aug 1, 2011 at 7:17 PM, John Tromp john.tr...@gmail.com wrote: On Mon, Aug 1, 2011 at 12:35 PM, steve uurtamo uurt...@gmail.com wrote: but they might be most fair. Only when there is ample evidence that the integer komi is in fact the perfect-play true komi. By definition of the scoring procedure the perfect-play komi must be integer. The only question is *which* integer? Due to the low probability of final positions with seki and an odd number of neutral (non-scoring) intersections, for area scoring (which is used at the olympiad) even number komi is far less likely to be optimal on an odd-surface board size (so this pretty much rules out 6.0 and 8.0). I'd say we have such evidence for 7x7, but certainly not for 9x9. There is of course no proof yet, but all statistics are IMO quite clearly pointing in the direction of 7.0 as the optimal komi; it is certainly the one that currently provides the most balanced winning ratio. (and btw, once we get that proof there really isn't much point in having such 9x9 tournaments any more anyway, right?). Perhaps needless to say, but I'm also in favor of lowering the komi. I'd really love to see a return to 6.5 komi combined with territory scoring (Steenvreter has supported that for years but I never had the chance to try it in a real tournament). However, given the difficulties some may have implementing Japanese style rules, I'd be quite happy with 7.0 komi also. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] EGC computer-go tournament 9x9
2011/8/1 Andrés Domínguez andres...@gmail.com: 2011/8/1 Erik van der Werf erikvanderw...@gmail.com: I'd really love to see a return to 6.5 komi combined with territory scoring (Steenvreter has supported that for years but I never had the chance to try it in a real tournament). However, given the difficulties some may have implementing Japanese style rules, I'd be quite happy with 7.0 komi also. Why not use chinese style rules but giving an extra point to white if black plays one move more? That would do the trick. It's a bit more subtle than that (e.g., think of mid-game passes, handicap stones, points in seki, game vs confirmation phase, etc.), but roughly speaking, yes, that does the trick. It is certainly feasible to 'translate' and area score to a territory score (in fact that is what I do in my program). However, when talking about the rules it just seems so much more natural to just describe the normal procedure for territory scoring :-) Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Tilburg 2011: list of hotels and schedule
yes On Fri, Jul 29, 2011 at 4:15 PM, Richard Lorentz lore...@csun.edu wrote: Does anyone else only see a screen of random characters when clicking on the list of hotels link? On 07/28/2011 11:31 AM, Rémi Coulom wrote: Hi, I updated the web site with new information: http://www.grappa.univ-lille3.fr/icga/event.php?id=43 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] computer-go tournament in Bordeaux in early august
On Mon, Jul 11, 2011 at 9:59 AM, Ingo Althöfer 3-hirn-ver...@gmx.de wrote: However, in go bots were also not trained especially for playing random positions. So, there is a difference. Monte Carlo bots play random positions all the time :-) Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] 19x19 opening books
Hi Aja, On Mon, Jul 11, 2011 at 4:47 PM, Aja ajahu...@gmail.com wrote: Hi Erik, Without a book Steenvreter usually still plays decent opening moves because it is pretty good at predicting pro moves (predictions are used to set priors in the tree). However, with longer thinking times there is sometimes a tendency for its style to become more cosmic (not sure if that is good or bad). On 9x9 or 19x19 do you mean? I used to have a different predictor for small boards, but now I just use the one trained on 19x19; it also works quite well on 9x9. Actually I was quite surprised, some years ago, when I found that something trained on 19x19 made the 9x9 program stronger... Of course for 9x9 the move predictor alone is not good enough. Sadly, to win tournaments it seems having a good book is now a must. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Zen!
Using a hashing scheme works perfectly if you can encode all relevant situational properties. Whether that's practical depends on the rules. I found that for the traditional rules it is generally feasible to encode all relevant situational properties. In my experience iterative deepening alpha-beta (when combined with an incomplete hash) has much more problems with graph-history-interaction than the usual MCTS implementations. This is because MCTS always plays out the sequence and therefore will discover any rule violation along the path, while alpha beta might make cutoffs that depend on unverified continuations. Once you start to update along multiple paths, propagating back to more than one parent, this may change of course. However, there are more fundamental problems with that idea anyway. E.g., if you update along some virtual path the statistics no longer necessarily conform to you your selection policy. Also, you have to avoid double counting when paths merge back. Erik On Sat, Jul 9, 2011 at 10:21 AM, valky...@phmp.se wrote: Yes it is tricky I think my main idea (and I think similar ideas can been found in a couple of papers) is to check for unexpected super ko violations and then one has to mark the path that led to that violations as dirty and create a new line in the collision node so that dirty variations has there own node linked to the original transposition node in a linked list, and there is a parent pointer that is used to identify the right node in the linked list. This is what I remember but details were hairy because of the many ways this could happen. But most of the problems goes away if one uses a good hashing scheme that also makes different capture histories different. I think Erik vand Der Werf has written some good stuff on this issue because it becomes extremely import when you try to solve game on small boards. -Magnus Quoting Michael Williams michaelwilliam...@gmail.com: Keeping a real tree is of course tivial. I guess you mean a way to preserve the benefits of transposition while also maintaining admisibility. That does seem like it would be tricky. On Fri, Jul 8, 2011 at 9:24 PM, valky...@phmp.se wrote: Quoting Michael Williams michaelwilliam...@gmail.com: The Valkyria tree is not a pure tree, because a node can have several parents if more than one sequence leads to a position. Best Magnus I think this is common, but inadmissable in the strictest sense, right? Because the optimal action for a node depends on it's history of positions thanks to the super ko rule. What was the word Don used for techniques like this? I mean techniques that are not going to lead to perfect play given infinite time and memory. Yes, you are right. For the next rewrite of Valkyria I actually think i rediscovered some algorithm to solve this but it is painfully complicated to implement. Luckily it is extremely rare that affect play (I think). __**_ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/**mailman/listinfo/computer-gohttp://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] Congratulations to Zen!
On Thu, Jul 7, 2011 at 9:01 PM, Ingo Althöfer 3-hirn-ver...@gmx.de wrote: Aja ajahu...@gmail.com ManyFaces (White) still wins even if it loses the semeai but captures the J4 group of 3 stones. And it's not possible for Black to save the J4 group and win the semeai at the same time. Yes, that is exactly what the experts in the forum found. The question: Do the other bots see this as well, and how early in the game do they see this outcome? Sure, its easy for strong bots :-) There are several ways to win in this position; no need to win the semeai. Even the J4 block can be traded for something like 2 points elsewhere (e.g., g4 a6). To win Black should have played 31 at 34. I think Nick's suggestion (between 32 and 35) is bad because the White group in the upper right corner is already dead. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Zen!
Nicks suggestion might work also, I'm not sure, but I had a lot more difficulties with it (against Steenvreter). With black 31 at e6 it all seems rather straightforward. I guess some programs may have difficulties seeing the corner is dead. Erik On Thu, Jul 7, 2011 at 10:54 PM, David Fotland fotl...@smart-games.com wrote: ManyFaces agrees with Nick that connecting on the right is better than 31 (with about 48% win rate). It doesn’t like moving at 34, as Erik suggests. After 34 (E6), it thinks white can win (55%) at H3, but this is too optimistic. David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Erik van der Werf Sent: Thursday, July 07, 2011 12:24 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Congratulations to Zen! On Thu, Jul 7, 2011 at 9:01 PM, Ingo Althöfer 3-hirn-ver...@gmx.de wrote: Aja ajahu...@gmail.com ManyFaces (White) still wins even if it loses the semeai but captures the J4 group of 3 stones. And it's not possible for Black to save the J4 group and win the semeai at the same time. Yes, that is exactly what the experts in the forum found. The question: Do the other bots see this as well, and how early in the game do they see this outcome? Sure, its easy for strong bots :-) There are several ways to win in this position; no need to win the semeai. Even the J4 block can be traded for something like 2 points elsewhere (e.g., g4 a6). To win Black should have played 31 at 34. I think Nick's suggestion (between 32 and 35) is bad because the White group in the upper right corner is already dead. Erik ___ 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] A Linear Classifier Outperforms UCT on 9x9 Go
On Wed, Jun 29, 2011 at 10:26 PM, Peter Drake dr...@lclark.edu wrote: On Jun 29, 2011, at 12:14 PM, Erik van der Werf wrote: I hope you are aware that some strong MCTS programs use (at least) a factor hundred less playouts to break even with gnugo. In fact, to get to 50% they don't even need a tree at all... (so UCT is perhaps not really that relevant at these levels) Yes -- as stated, this new method is not (yet?) competitive with cutting-edge MCTS (e.g., RAVE and fancy domain-specific playouts). Our claim is merely that it beats vanilla UCT. That claim can't be true in general. Sure you can beat a weak implementation at some fixed playout level but in general (e.g., as the number of simulations grows much larger) its simply not possible because the linear classifier alone doesn't have the capacity to represent a big tree (which eventually it must be able to do to infer deep tactics). Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] A Linear Classifier Outperforms UCT on 9x9 Go
On Wed, Jun 29, 2011 at 7:17 PM, Brian Sheppard sheppar...@aol.com wrote: Why is a classifier better than having a lookup table indexed by OurLastMove, OppLastMove, ProposedNextMove that returns the Wins / Trials experienced when ProposedNextMove is played after the sequence OurLastMove, OppLastMove? Just my guess, but probably it's because the shared bias provides more generalization, and perhaps there's some benefit from having a forgetting factor. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Zen19D is blitz 5d in KGS
On Wed, Jun 1, 2011 at 3:29 PM, Petr Baudis pa...@ucw.cz wrote: Hi! On Wed, Jun 01, 2011 at 11:52:07AM +0800, Aja wrote: Zen19D, running on a small cluster with 26 cores, is rated 5d in KGS now. I played several games with it and felt Zen is getting close to my level. I am happy to see that Crazy Stone and Zen are both so strong. Looks like in computer Go community we got 1-2 stones improvement in the past year. Yes, I just wish people would also publish their improvements. ;-) My guess is that the main improvement of Zen and CrazyStone is in better domain-specific tactical heuristics? Do you really think there have been big unpublished improvements recently? Several programs already got to 3d/4d (kgs blitz) some time ago on much smaller machines. My guess would be that this recent increment to 5d is primarily due more powerful hardware... Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Pachi Authors
On Mon, Apr 18, 2011 at 5:49 PM, Jean-loup Gailly jl...@gailly.net wrote: The actual value of one handicap stone is twice the normal komi so about 15 points(*) Perhaps 15 (per additional handicap stone) would be ok if you had pro-level playouts. The komi points for White are 100% safe, while Black's extra stones on the board can all still die easily in weak playouts... Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Pachi Mailing List
On Sun, Apr 17, 2011 at 2:58 PM, Fuming Wang fuming...@gmail.com wrote: (Also, we have finally set up a real (though simple) homepage for Pachi at http://pachi.or.cz/.) Happy go research, According to information on Pachi's homepage, Pachi won a 7h game against Zhou Junxun 9p, who is an active 9p player and has won at least one world title. I would say this put Pachi at top amature or beginning pro level. On the other hand, Pachi is around 3d on KGS, which I am not very familar, but I am guessing that it only proves that Pachi is at about median amature level. What's your assessment, and thinkings about this difference (if there is a difference in your assessment)? Pro ranks are spaced differently (1p~=7d ,..., 9p ~= 9d). It is not strange for a 3d amateur to win with 7 stones handicap. E. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] living by two false eyes
On Wed, Apr 6, 2011 at 4:11 PM, fastgo fas...@gmail.com wrote: In StoneGrid, it has built in the Benson algorithm in the playout. It does not seem to help much on overall strength. But I think it should solve the case here. I think you can implement the Benson Algorithm in the late stage of the playout. It has another advantage finishing the playout in shorter length. In StoneGrid, the playout on average finishes in a little bit more than 70 stones on 9x9. You need quite a lot of effort to get it fast as well as correct. So the return of the effort is low. I also have it in my program. The funny thing is that if I disable it the program actually becomes slower because of the longer playouts... Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] living by two false eyes
On Wed, Apr 6, 2011 at 6:05 PM, Brian Sheppard sheppar...@aol.com wrote: How does the Benson algorithm result in shorter playouts? The playout can terminate early when you have a proof that more than half of the points belong to one side (my program not only determines life but also territory). Also, as others already pointed out, avoiding bad moves tends to shorten the playouts. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Where are the Jigo(lo)s?
On Mon, Apr 4, 2011 at 10:50 AM, Aja ajahu...@gmail.com wrote: Maybe it is because the programs are not so close in playing strength. Anyway, it is the first KGS tournament of komi 7.0 so maybe most of the participants didn't prepare well. As far as I know, ManyFaces even didn't handle Jigo specially. Also, Steenvreter played many games using a BAD book (built for 7.5 if I remember correctly) and finally Erik removed the book completely. Actually I changed book twice. The first one was built for 7.0 but after 10 or so games it was clear that some lines were too bad against strong opponents. I think the reason was a combination if it being auto-generated (no human inspection) and not being deep enough (it had only run for two days). Then I shortly tried an old book, made for 6.5 komi, but didn't like that either so the second half of the tournament I just ran without book. BTW in some recent self-play test games Steenvreter got around 10 to 15% jigo's. I guess the reason why there were fewer in the tournament is because jigo's are less likely when the programs are more different in strength. Best, Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] 7.0 Komi and weird deep search result
Even without exploration. With N going to infinity and, I suppose, the implicitly assumed infinite memory to store results with infinite precision, thus eventually representing the complete tree, all you have to do is not expand solved branches. I don't see how that would be different for 3-valued spaces. Anyway, that's just plain old minimax; such arguments probably have very little to do with actual performance. I don't know about other programs but Steenvreter does have an exploration term. My impression is that ignoring the fact that for integer komi the distribution is not binomial makes it explore too much rather than too little. Erik On Mon, Apr 4, 2011 at 2:01 PM, Michael Williams michaelwilliam...@gmail.com wrote: Yeah. As N goes to infinity, MCTS+RAVE goes to MCTS. So it sems like it would be guaranteed to converge only if you had an exploration term in the MCTS. On Sun, Apr 3, 2011 at 10:53 PM, Petr Baudis pa...@ucw.cz wrote: On Sun, Apr 03, 2011 at 04:21:10PM -0400, Brian Sheppard wrote: AIUI, RAVE without special modifications (like those done in Mogo later) does not have any convergence guarantees either. MCTS using RAVE prioritization *does* converge to game theoretic values in a binary-valued space. I do not think this is true, at least for the general space of simulation functions, as they may entirely shun the winning move. Can you reference some more detailed analysis claiming this? -- Petr Pasky Baudis UNIX is user friendly, it's just picky about who its friends are. ___ 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] April KGS bot tournament: 9x9, fast
I think a human challenging the program can set this manually. It would be nice though if we could define the preferred setting with something like rules.komi=x. Erik On Wed, Mar 30, 2011 at 3:27 PM, valky...@phmp.se wrote: Is there a way of testing with komi 7 on KGS? As far as I know the mode=custom setting does not allow for setting the komi, or it is not documented. Magnus Quoting Nick Wedd n...@maproom.co.uk: I had forgotten about the poll. In fact a clear majority voted for a komi of 7. I have changed the tournament - it will use komi of 7, not 7.5. ___ 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] UCT parameters and application to other games
It sounds like you're using a classical (deterministic) evaluation function. Try combining UCT with Monte Carlo evaluation. Erik On Sat, Mar 26, 2011 at 12:43 PM, Daniel Shawul dsha...@gmail.com wrote: Hello, I am very new to UCT, just implemented basic UCT for go yesterday. But with no success so far for GO,I think mostly because it searches not very deep (depth = 3 on a 5 sec search with those values). I am using the following values as UCT parameters UCTK = sqrt(1/5) = 0.44 UCTN = 10 (visits afte which best move is expanded) Even if I lower UCTK down to 7 I get a maximum depth of d=7 at the start position for a 5 sec search. For how deep a search should I tune these parameter for ? Before UCT, I have an alpha-beta searcher which sometimes plays on CGOS. It reached a level of ~1500, and this engine seems to be too strong for the UCT version. It just gets outsearched in some tactical positions and also in evaluation I think. For example, I have an evaluation term which gives big bonuses for connected strings which seems to give an edge in a lot of games.. How do you introduce such eval terms in UCT ? But for my checkers program , to my big surprise , UCT made a significant impact. The regular alpha-beta searcher averages a depth=25 but the UCT version I think is equally strong from the games I saw. That was a kind of surprise for me because I thought UCT would work better for bushy trees and when the eval has a lot of strategy. It also reached good depths averaging 16 plies . My checkers eval had only material in it, so I don't know if UCT is bringing strategy (distant information) to the game which the other one don't have.The games are not really played out to the end rather to a MAX_PLY = 96 afte which the material is counted and a WDL score is assigned (I call it partial playout). Also the fact that captures are forced seem to help a lot because it doesn't make too many mistakes. I also found out some positions where it encounters similar problems as ladders in go. But in the checkers case, this problems are still solved correctly. Only problem is that it doesn't report correct looking winning rates. For example, in a position with two kings where one of the kings is chasing the other to the sides to mate it, but the loosing king can draw by making a serious of correct moves to get itself to one of the safe corners; The program displays winning rates of 0.01 (when it should have been more like 0.5) but it still manages the draw ! thanks and apologies for the verbose email Daniel ___ 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] UCT parameters and application to other games
I agree, if you use a hard limit it should be much higher (probably something like twice the board surface is ok). 110 moves is just an observation of the average playout length for the empty 9x9 board. With smarter playouts that average tends to become lower, but the distribution may still have a long tail. Erik On Sat, Mar 26, 2011 at 3:36 PM, Rémi Coulom remi.cou...@free.fr wrote: I'd recommend more than 110. Maybe 200 is better. In Crazy Stone, I use no limit, and test for superko. Rémi On 26 mars 2011, at 15:32, Daniel Shawul wrote: Sorry 81 moves was a bad estimate by me. I am actually using 96 moves. I will change that to 110 or above moves and see what effect it has. Also I would take Remi's suggestion i.e to bias the move selection process. For the alpha-beta program , I have a decent move ordering algorithm and qsearch. I guess can borrow some from that. In the meantime, I found a paper using UCT for chinese checkers and other games http://www.google.com/url?sa=tsource=webcd=7ved=0CD4QFjAGurl=http%3A%2F%2Fweb.cs.du.edu%2F~sturtevant%2Fpapers%2Fmpuct_icga.pdfrct=jq=UCT%20for%20checkersei=mPiNTcqdBsSO0QG20-nWCwusg=AFQjCNGFgMMMG8xMtawvx-3rQtwXPhfWxQcad=rja, and also some fun java programs using UCT for checkers. It seems UCT is indeed competitive in checkers. I must say I didn't expect that at all. I think the forced nature of captures helps to improve tactical awareness of the MC simulations. Is that so ? On Sat, Mar 26, 2011 at 8:52 AM, Erik van der Werf erikvanderw...@gmail.com wrote: Ah ok, I misunderstood. Still something seems to be wrong. On the empty 9x9 board I think most programs with random/light playouts play in the order of 110 moves. ~81 moves seems quite low; in my experience you can only get such low numbers to work well if you have a lot of knowledge in your playouts. Did you check the quality of the evaluations/playouts? If you want UCT to search deeper you need good priors and perhaps something like rave/amaf. Best, Erik On Sat, Mar 26, 2011 at 1:13 PM, Daniel Shawul dsha...@gmail.com wrote: Hello, I am using monte carlo playouts for the UCT method. It can do about 10k/sec. The UCT tree is expanded to a depth of d = 3 in a 5 sec search, from then onwards a random playout (with no bias) is carried out. Actually it is a 'patial playout' which doesn't go to the end of the game, rather upto a depth of MAX_PLY=96. If the game has ended earlier, then a win/draw/loss is returned. Otherwise I forcefully end the game by using a determinstic eval and assign a WDL. For 9x9 go actually most of random playouts end before move 81. For the alpha-beta searcher , I do classical evaluation. With heavy use of reductions I can get a depth of 14 half plies , which seems to give it quite an edge against the UCT version. Is the depth of expansion for the UCT tree too low ? (d = 3 in a 5 sec search). Should I lower the UCTK parameter to 0.1 or so which seems to give me a depth = 7 at the start positon of a 9x9 go. I am confident my implementation is correct because it is working quite well in my checkers program despite my expectation. thanks Daniel On Sat, Mar 26, 2011 at 7:54 AM, Erik van der Werf erikvanderw...@gmail.com wrote: It sounds like you're using a classical (deterministic) evaluation function. Try combining UCT with Monte Carlo evaluation. Erik On Sat, Mar 26, 2011 at 12:43 PM, Daniel Shawul dsha...@gmail.com wrote: Hello, I am very new to UCT, just implemented basic UCT for go yesterday. But with no success so far for GO,I think mostly because it searches not very deep (depth = 3 on a 5 sec search with those values). I am using the following values as UCT parameters UCTK = sqrt(1/5) = 0.44 UCTN = 10 (visits afte which best move is expanded) Even if I lower UCTK down to 7 I get a maximum depth of d=7 at the start position for a 5 sec search. For how deep a search should I tune these parameter for ? Before UCT, I have an alpha-beta searcher which sometimes plays on CGOS. It reached a level of ~1500, and this engine seems to be too strong for the UCT version. It just gets outsearched in some tactical positions and also in evaluation I think. For example, I have an evaluation term which gives big bonuses for connected strings which seems to give an edge in a lot of games.. How do you introduce such eval terms in UCT ? But for my checkers program , to my big surprise , UCT made a significant impact. The regular alpha-beta searcher averages a depth=25 but the UCT version I think is equally strong from the games I saw. That was a kind of surprise for me because I thought UCT would work better for bushy trees and when the eval has a lot of strategy. It also reached good depths averaging 16 plies . My checkers eval had only material in it, so I don't
Re: [Computer-go] I need an off-the-shelf final position live/dead evaluator
AFAICS this (japanese-test-7.sgf) is simply not a final position. If white plays first at least there's a ko and he will probably capture a lot of black stones. Moreover, I think officially the group would be in seki if the dame are not filled (so then none of the intersections around S6 are territory). Erik Of course Black's position is hopeless so he might as well have resigned. On Sun, Feb 20, 2011 at 8:06 PM, Rémi Coulom remi.cou...@free.fr wrote: On 20 févr. 2011, at 19:53, Nick Wedd wrote: Crazy Stone is probably wrong, too, because there are some complicated things happening around S6. I would say S6 is dame, not S7 and S5. B12 and P1 look like really weird dame, though. So do K15 and N12. The right-side of the board is complicated, but the rest should be obvious for GNU. Unless I am mistaken... Something must be wrong. There are problems in this position. But S6 is not one of them, it is very clearly black territory. Maybe you have the wrong SGF files, or they have been rotated? Nick After Black connects all the false eyes in the top right, and White fills dame, then White could kill by atari at T6. So I believe Black must play S6 to prevent T6, so S6 is dame. I am really having a lot of fun with Japanese rules :-) 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
Re: [Computer-go] Could a 'doubling dice'** encourage early resignation by programs?
I once played in a tournament like that. What happened was that the stronger player would typically double very early in the game (e.g. on the first dubious non-joseki move). In one game my opponent was 6d (~500 Elo stronger than me) so when he increased the stakes I immediately resigned; I probably would have resigned even if he had doubled on move 2. Is this really what you want? Erik On Thu, Jan 27, 2011 at 3:31 PM, Jonathan Chetwynd j.chetw...@btinternet.com wrote: Could a 'doubling dice'** encourage early resignation by programs? each program would have to forfeit a double game, if it played on and lost the game, but could resign for a single loss. scores in earnest might need to be tallied in the public arena. though one would hope that the application designers. regards Jonathan Chetwynd http://www.peepo.com **as per backgammon, either player can double, but then it is the opponent's choice to resign or accept, and to redouble at their discretion, though this aspect may be ott. apologies if I missed the obvious, it seems I omitted the most severe error, playing on, when a game is lost. in the thread: Are 4 'easy to avoid errors' common to all MC programs? ___ 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] Beta-testing: feedback to bot owners
On Sat, Jan 22, 2011 at 2:53 PM, Nick Wedd n...@maproom.co.uk wrote: If I insist on running events with integer komi, I know what will happen. Some bots, including GNU Go, already support it; some will implement it correctly; some will implement it wrong, so that strange things happen; some will fail to support it, and thereby lose won games to weaker programs; some may refuse to support it, and stop playing in the events that I organise. I prefer to leave things as they are. Perhaps, or one could consider it a nice incentive to fix the issue. Maybe I missed it, but has anyone actually announced that they would stop participating because of integer komi? I announced earlier that I would be using integer komi of 7 for the 9x9 KGS bot tournaments this year. I have changed my mind, I will use half-integer komi throughout. This is not an ideal decision, it is a pragmatic one. Too bad, I think 7 komi would have been interesting. I suppose there is no hope of persuading you to try 6.5 komi + Japanese rules... :-) Best, Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Combinatorics of Go
Hi Robert, Perhaps my answer was a bit cryptic. I'll try to explain. In a computer go program it is indeed needed to detect cycles when you want to claim, for example, a tie or no-result. So you're right about that. However, to evaluate a position and infer the best move it is generally not needed as long as you don't care about distinguishing between not having a final result due to the position being too complex to reach a terminal state and not having a final result because the search will never reach a final state. (Notice though that in both cases eventually the optimal move can still be found). Say for simplicity we select our move based on an iterative deepening alpha beta search that ignore some aspects of the history. In case of a potential long cycle the search will simply not return a final result unless the komi is high or low enough for one side to deviate from the cycle. The way I think about it is that a cycle is only a true cycle when nothing in the state distinguishes the first occurrence of a position form later occurrences. Superko requires adding something to the state that distinguishes the first occurrence from later occurrences, so in that sense I meant that the cycle is no longer a true cycle (i.e., it is broken). Regarding the conjecture you propose, I'd rather formulate it as: Under Japanese style ko rules, the long-term history is never needed to infer the best move. Some short-term history is however still needed for things like basic-ko end game-ending passes. Best, Erik On Mon, Jan 3, 2011 at 7:57 AM, Robert Jasiek jas...@snafu.de wrote: On 02.01.2011 22:04, Erik van der Werf wrote: to 'not return a result' you don't need the history. How? A cycle is a presupposition for the result No Result (or long cycle tie). (Of course, hashing by numbers of stones on the board or Cycle Law's prisoner difference etc. may often be sufficient, but if it is not, then determining a cycle by referring to history is necessary.) When optimal play leads to a cycle a) necessarily b) optionally then any state in that cycle should lead to an equivalent cycle. I am not sure exactly what you mean by this. Do you want to say that, given the game's move-sequence so far, its last moves, if they are about to be in cycles, then necessarily they are in the same cycle? You still need to establish the fact whether they are in a cycle. Superko rules, always returning a result, just need more because they break more. How do they break more? They break less by not creating any repetition:) Maybe you want to say: They have to detect the first repetition before it is executed. Ok, let us assume that under Japanese style rules the programs just recycle a bit, using the same or different cycles, until trying to apply the long cycle rule. They still need to detect that they have been playing some cycle at all. See above: You need the history because hashing does not guarantee detection. So, you or Olivier tell me why not having to use history! Are you having some insight I overlook? (In CG practice, the history can be ignored as a first approach - under superko or Japanese style ko rules. To be sure though, history is needed.) Do we need to do the maths? Your conjecture seems to be: Under Japanese style ko rules, the history is never needed to apply the No Result / long cycle tie rule. I claim the opposite (might be needed). -- robert jasiek ___ 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] Combinatorics of Go
On Mon, Jan 3, 2011 at 2:13 PM, Robert Jasiek jas...@snafu.de wrote: On 03.01.2011 13:44, Erik van der Werf wrote: This is handled trivially by observing that one sided passes/captures more in each cycle. How do you distinguish that from the opposing program passing as a tactical mistake (or as a psychological trick)? If the opponent passes this knowledge becomes available by the small finite list of extra features I described before. BTW I don't see how one could play a psychological trick against a program. I suppose the best one could hope for is to trigger a bug. can work out the details. Easy: almost never is a better description. The word 'almost' to me suggests that you would know for sure that there exists an exception. I'm not convinced that is the case. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] News on Tromp-Cook ?
On Sun, Jan 2, 2011 at 1:20 AM, Aja ajahu...@gmail.com wrote: Do you mean the whole MCTS scheme combined with UCB formula proposed by Mogo is completely inspried by Levente's work? I cannot speak for the Mogo team, but it is clear to me that the idea was simply 'out there' when Mogo started and they had access to it through the various research communities (Machine learning, Computer games, etc.). I guess at least one of their supervisors most have had some contacts with Levente or Csaba. Levente did not discuss UCB-tuned, so they might have gotten that directly from Auer. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go