[Computer-go] PhD Thesis - Lars / Gomorra
Hi all, after quite some years I eventually completed my PhD. As the thesis is too large to be accepted as attachment by this list, please find my thesis for download here (until I find a proper place for hosting it): https://www.dropbox.com/s/x0lh7ky5lvj6c1y/PhdThesisSchaefers.pdf Perhaps some of you will find some time to have a look into it. The largest part is about the parallelization of MCTS for HPC clusters that I developed and used with Gomorra. I am happy that the end of my PhD study is actually not the end for Gomorra. Tobias Graf started his PhD study some month ago and continues to work on/with Gomorra. He is a very talented guy and I'm happy that he is now pushing things forward. With the end of my PhD study I also left academia for now. But who knows how life goes on.. Thanks a lot to all of you! It was a lot of fun to be part of this friendly and constructive community. I will keep on reading this list and hope to see some of you in the future again! For those of you who like a more compact read of the parallelization part of my thesis: I will soon come along with a recently accepted T-CIAIG paper about it. Just have to check if the copyright terms allow me to distribute an author manuscript on this list. Best wishes, Lars -- Lars Schaefers Computer Engineering Group of Prof. Dr. Marco Platzner Paderborn Center for Parallel Computing, University of Paderborn Pohlweg 47-49, 33098 Paderborn, Germany Tel: +49 (0)5251 60 4341, Fax: +49 (0)5251 60 5377 Office: Building O 3.119 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Crazy Stone!
> I don't think it is wrong. I do think it is inadvisable. You are > trusting the server's score-counter to count in the way that you > expect, when there are dead groups on the board and it can reasonably > assume that there are no dead groups on the board. > > For instance, in the position shown, it counted B2 as a point for > White, even while it was regarding the black stones at C1 and C3 as > alive. Did your code _know_ it was going to do that? I think the scoring rules in the clean-up phase are pretty clear (and even simpler than in the normal playing phase). After two consecutive passes in the clean-up phase it says that all stones are regarded as alive. It is just dangerous to continue play only to further clarify the situation. I think it's more likely that you might lose on time because of a network lag or some yet undiscovered bug in your engine (or the server/gtpClient) than that you might lose because the server will count wrong. Especially when playing with a cluster there are plenty of possibilities to crash :/ This is at least the reason why Gomorra also passes as early as possible, when the game is won following the KGS rules. And in the past the tournament report often said about such games that Gomorra failed to correctly cleaning up the board, but won anyway. This was for the reason given above. Congratulations to CrazyStone! Thank you Nick for the report! - Lars -- Lars Schaefers Computer Engineering Group of Prof. Dr. Marco Platzner Paderborn Center for Parallel Computing, University of Paderborn Pohlweg 47-49, 33098 Paderborn, Germany Tel: +49 (0)5251 60 4341, Fax: +49 (0)5251 60 5377 Office: Building O 3.119 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] On Semeai Detection. Was: Zen resignation positions
Hi Robert, I too have problems with receiving mails form this list. Hence, sorry for late replies. Thanks to veg for working on this issue! 1) Use of the term semeai In this paper the term "semeai" is indeed not used in the most accurate fashion. We realized this only after the paper was already accepted (thanks to a comment by Rémi Coulom) and addressed this issue in the conclusion section: "For the sake of correctness, we must admit that the term semeai might not be completely appropriately used throughout this paper. A score cluster does not always need to be caused by the presence of a capturing race. In any case, it represents an evaluation singularity that is likely caused by uncertainty in the evaluation of the life and death state of one or more groups of pieces. Also in this case our approach will help to localize the respective groups on the board." 2) Identification of score clusters The score clusters are obtained by analyzing the observed outcomes of the semi-randomized MC simulations. The identification of the critical board regions has no two-value-logic evidence. 3) two-safe-group It is a term used by Aja Huang to describe a class of Go positions he created, that contain two safe but not yet completely settled groups. The positions were created with the aim to be difficult to understand for current MC programs. Aja Huang and Martin Müller recently published a very nice paper about this test suite and experimental results with a number of strong Go programs. You can find their paper "Investigating the Limits of Monte Carlo Tree Search Methods in Computer Go" here: http://webdocs.cs.ualberta.ca/~mmueller/publications.html 4) runtime We plan to use this heuristic foremost in near root nodes. It is certainly to expensive to be used all over the tree. The collection of the necessary data is pretty efficient. The mode seeking part is a bit more involved, but not more expensive than the computation of other state-of-the-art heuristics used for move prediction. This is at least the case for Gomorras implementation of those heuristics. Once again: The described heuristic is yet not used in our "productive" version of Gomorra. Its application to improve the MCTS search process is still research in progress. Cheers, Lars On 09.10.2013 15:30, Lars Schäfers wrote: > "On Semeai Detection in Monte-Carlo Go" In this paper, does "semeai" stand for "one of 'capturing race', 'unsettled life and death situation', 'basic ko fight', 'multiple ko fight', 'ladder breaking' 'etc.'"? Is the identification of a cluster proven only with stochastic probability, or is there even two-value-logic evidence? What is a two-safe-group? How do the methods affect simulations runtime? -- robert jasiek -- Lars Schaefers Computer Engineering Group of Prof. Dr. Marco Platzner Paderborn Center for Parallel Computing, University of Paderborn Pohlweg 47-49, 33098 Paderborn, Germany Tel: +49 (0)5251 60 4341, Fax: +49 (0)5251 60 5377 Office: Building O 3.119 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] On Semeai Detection. Was: Zen resignation positions
Hi, partly as a result of the discussion in the thread "Zen resignation positions", we wrote the following paper on the detection of semeai during MCTS: "On Semeai Detection in Monte-Carlo Go" Abstract: A frequently mentioned limitation of Monte-Carlo Tree Search (MCTS) based Go programs is their inability to recognize and adequately handle capturing races, also known as semeai, especially when many of them appear simultaneously. This essentially stems from the fact that certain group status evaluations require deep lines of correct tactical play that somewhat oppose to the exploratory nature of MCTS. In this paper we provide a technique for heuristically detecting and analyzing semeai during the search process of a state-of-the-art MCTS implementation. We evaluate the strength of our approach on game positions that are known to be difficult to handle even by the strongest Go programs to date. Our results show a clear identification of semeai and thereby advocate our approach as a promising heuristic for the design of future MCTS simulation policies. You can download an author manuscript of the paper here: http://www.cs.uni-paderborn.de/fachgebiete/computer-engineering-group/people/schaefers.html Cheers, Lars On Tue, 2012-12-11 at 22:02 +, "Ingo Althöfer" wrote: > Kahn Jonas > > ... and that's not interesting: We want to focus on the multimodularity. > > > > So just count the number and depths of peaks. > > Jonas is right. > Identifying peaks and their "volumes" is indeed rather easy. > > For the long run I see a plan with two stages. Stage (ii) should be > interesting particularly as along as pro players still give handicap > stones to bots. > > (i) Given a large board (typically 19x19) and for some middle game position > a Monte-Carlo histogram with two or more peaks. Identify the corresponding > local fight(s) which is/are responsible for the peak(s). This task is not > trivial, but within reach. > > (ii) When still ahead in such a position, the bot should "resolve" a local > fight for the prize of a "few" points. An example for this can be seen > in ds' and Petr Baudis' comments on a game between Catalin Taranu (5p) > (=egc2012pro) and Crazy Stone on August 02, 2012. > http://www.gokgs.com/gameArchives.jsp?user=crazystone&year=2012&month=8 > The upper left corner was no completely cleared, and CrazyStone played > with the resulting peaks for more than one hundred moves. In the > comments Petr tried to defend CrazyStone, but ds claimed that early > action by the bot would have avoided the mis-evaluations. > > ** > One of my hopes (in contrast to pessimists like Stefan Kaitschick) is: > When there are "MANY" unresolved local fights things might become less > complicated again for MC bots. (Side remark: on small boards positions > with several peaks are not seldom.) > > Ingo. > ___ > Computer-go mailing list > Computer-go@dvandva.org > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Lars Schaefers Computer Engineering Group of Prof. Dr. Marco Platzner Paderborn Center for Parallel Computing, University of Paderborn Pohlweg 47-49, 33098 Paderborn, Germany Tel: +49 (0)5251 60 4341, Fax: +49 (0)5251 60 5377 Office: Building O 3.119 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Go playing software accelerator development
Nice to see somebody working on accelerator hardware for Go (again). There actually were some attempts in the past and even publications that might be interesting for you [1][2]. I think in principle you have two possible directions when looking at accelerating basic MCTS for Go: a) Accelerating the computation of single playouts b) Computing a large number of playouts in parallel (likely using some pipelining technique) The problem with b) is that N parallel simulations are less valuable then N sequential ones, as simulations in MCTS are guided depending on former simulation outcomes. See e.g. [3] for according experimental results. You would likely end up with something that was called leaf-parallelization [4]. a) would be great, but is extremely difficult to achieve. The computation of strong playouts doesn't pose to much points for fine-grained parallelization and kind as well as amount of computations that need to be done for the generation and placement of moves is pretty inhomogeneous. Likely your hardware will run at a speed of about 150 to 200 Mhz (?), making it difficult to achieve a significant speedup over modern CPUs with 8 or more cores each running at more than 3 Ghz. I'm somehow convinced that looking at accelerator hardware for MCTS Go might be a bit to early. The algorithm and heuristics used are still changing a lot from year to year and adapting your HW design is likely more involved than adapting software. But that's just my personal point of view. Good luck! - Lars [1] Gao, Haiying, Wang, Fuming, Lei, Wei, and Lin, Yun: Monte Carlo simulation of 9x9 Go game on FPGA , IEEE International Conference on Intelligent Computing and Intelligent Systems, 865–869, October 2010 [2] Koizumi, Kenichi, Inaba, Mary, Hiraki, Kei, Ishii, Yasuo, Miyoshi, Takefumi, and Yoshizoe, Kazuki: Triple Line-Based Playout for Go - An Accelerator for Monte Carlo Go, Proceedings of the 2009 International Conference on Reconfigurable Computing and FPGAs, 161–166, 2009 [3] Segal, Richard B.: On the Scalability of Parallel UCT, International Conference on Computer and Games, 36–47, 2010 [4] Chaslot, Guillaume M.J-B., Winands, Mark H.M., and van den Herik, H. Jaap: Parallel Monte-Carlo Tree Search, Conference on Computers and Games, 60–71, 2008 On Wed, 2013-05-22 at 06:35 +, Рождественский Дмитрий wrote: > Incredible, 100 nanoseconds is only about 300 instructions of a CPU. > Are you talking about 19x19? And 1 microsecond for my design will > probably be a worst-case (as I calculate freedom and capture > iteratively). When almost all stones have free places around it will > be down to ~100 nanoseconds. > As to the number of possible accelerators on-chip - it varies upon > price. I think it can be 5-250, for the price $250-$5000. So the cost > of a single simple accelerator will be $20-$50. > > Dmitry > > 21.05.2013, 23:13, "Mark Boon" : > > Sounds interesting. But 1 microsecond for a move is not particularly > > fast. There are already implementations that do that in the 100-300 > > nanoseconds range on one core. 1 microsecond is probably considered > > as 'semi-light' playout. I suppose the question then becomes, how > > many of these could your accelerator do in parallel? > > > > Mark > > > > > > On Tue, May 21, 2013 at 8:06 AM, Alexander Kozlovsky > > wrote: > > Я тоже кстати из ЛИАПа, с четвертого факультета, может и > > пересекались :) > > > > > > On Tue, May 21, 2013 at 7:02 PM, Рождественский Дмитрий > > wrote: > > Hi all, > > > > I have got an idea to create a hardware accelerator > > for Go playing software. It will probably be a USB > > (or maybe PCI-Express) device that will be able to > > do some basic, but very time-consuming for > > general-purpose CPU calculations very fast. For > > example load a goban layout, make a number of random > > moves (as used in Monte-Carlo algorithm) and unload > > result back to a computer. > > > > As long as it will be a hardware, it will be able to > > do specified calculations only, but the speed will > > be very high. For example, making just a copy of the > > particular goban layout will require typically about > > 10 nanoseconds only (one internal clock cycle). > > Calculation of the validity and results of a > > particular move (including a check for ko and > > captured stones) will probably take 1 microsecond. > > This as usual may vary during debugging, but the > > current move calculation engine draft I've started > > to develop is about this figures. > > > > My nearest aims here are: > > -
Re: [Computer-go] Anomalies in MCTS
Gomorra also gets it right. It uses RAVE and also has seki knowledge in the playouts. - Lars On Tue, 2013-02-05 at 05:50 +, David Fotland wrote: > Many Faces of Go gets this one right (my current unreleased engine, not the > 12.022 download). It requires some seki knowledge in the playouts. > > David > > > -Original Message- > > From: computer-go-boun...@dvandva.org [mailto:computer-go- > > boun...@dvandva.org] On Behalf Of "Ingo Althöfer" > > Sent: Monday, February 04, 2013 5:38 AM > > To: computer-go@dvandva.org > > Subject: Re: [Computer-go] Anomalies in MCTS > > > > Hi Cameron, hi Jonas, > > > > thanks for your feedback. > > > > > You might try some of the RAVE counter-examples, such as the following > > > position from Martin Mueller (White to move): > > > http://www.cameronius.com/research/go-rave-blunder-1.pdf > > > > > > It's a seki position in which FUEGO chooses the correct move B2 > > > without RAVE, but the incorrect move D9 with RAVE enabled. > > > > I translated this position in sgf. Programmers can find it at > > www.althofer.de/mueller-rave-blunder-1.sgf > > and may try with their bots. > > > > Regards, 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 -- Lars Schaefers Computer Engineering Group of Prof. Dr. Marco Platzner Paderborn Center for Parallel Computing, University of Paderborn Pohlweg 47-49, 33098 Paderborn, Germany Tel: +49 (0)5251 60 4341, Fax: +49 (0)5251 60 5377 Office: Building O 3.119 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] [Off-topic] crowdfunding a go promotion
This is a bit off topic (sorry), but might be of interest for some of you. I'll keep it short.. A colleague of mine and his brother just started a crowdfunding project on indiegogo with the goal to collect some money for the production of a Go promotion video. In case of interest have a look here: http://www.indiegogo.com/playmorego otherwise ignore it and sorry for bothering you.. Lars ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Recursive Neural Networks
Thank you Simon for pointing to our paper. I just would like to send you the link below to the actual camera ready version of our paper (as it was published) with some modifications related to comments from the reviewers and some feedback kindly provided by Rémi Coulom and Ruby Chiu-Hsing Weng: http://typo3.cs.uni-paderborn.de/fileadmin/Informatik/AG-Platzner/publications/schaefers12_cig/schaefers12_cig.pdf Lars Schaefers On Thu, 2013-01-24 at 14:54 +, Lucas, Simon M wrote: > Nice paper in IEEE Conference on Computational Intelligence and Games 2012 > that compares Remi's method with Stern et al and some others: > > "Comparison of Bayesian Move Prediction Systems for Computer Go" > Martin Wistuba, Lars Schaefers, and Marco Platzner > > The paper is normally available for free download from the CIG 2012 web site: > http://geneura.ugr.es/cig2012/ > > but the server seems to be down at the moment so I've just put it here > temporarily: > > http://dces.essex.ac.uk/staff/sml/tmp/cig2012MovePredictionGo.pdf > > Simon Lucas > > > > -Original Message- > From: computer-go-boun...@dvandva.org > [mailto:computer-go-boun...@dvandva.org] On Behalf Of Petr Baudis > Sent: 24 January 2013 14:30 > To: computer-go@dvandva.org > Subject: Re: [Computer-go] Recursive Neural Networks > > On Wed, Jan 23, 2013 at 04:41:57PM -0500, George Dahl wrote: > > This paper reports 36% move prediction accuracy: > > http://www.cs.utoronto.ca/~ilya/pubs/2008/go_paper.pdf > > C.f. also http://research.microsoft.com/apps/pubs/default.aspx?id=67955 > which reports 34% accuracy for top move, 66% accuracy for top five moves > suggested. > > I'm not sure if anyone measured Remi Coulom's pattern model performance in > move prediction. > > P.S.: As already mentioned, it should go without saying that there is very > little correlation if any between move prediction rate and playing strength > as a sole move generator or a feature provider. > > Petr "Pasky" Baudis > ___ > Computer-go mailing list > Computer-go@dvandva.org > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go > ___ > Computer-go mailing list > Computer-go@dvandva.org > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Lars Schaefers Computer Engineering Group of Prof. Dr. Marco Platzner Paderborn Center for Parallel Computing, University of Paderborn Pohlweg 47-49, 33098 Paderborn, Germany Tel: +49 (0)5251 60 4341, Fax: +49 (0)5251 60 5377 Office: Building O 3.119 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Should playout patterns be gathered from high-dan games only?
Hi Alexander, for the reasons you mentioned I trained my pattern system on games between weaker and stronger players some years back. That was for 9x9 board size where I used game records from cgos where both players had a strength of at least 1700 Elo and at least one player was stronger than 2400 Elo. Obviously you should only learn the moves of the players that are stronger than 2400 ELO here.. I got some significant improvements of it that time, although I can't provide numbers today. I didn't made comparable experiments for larger board sizes. - Lars On Wed, 2013-01-16 at 15:53 +, Alexander Kozlovsky wrote: > Hi everybody! > > Is it enough to use big number of hign-dan games > to build a strong pattern library for random playouts? > > I have suspiction the most useful patterns may be > gathered from low-dan games. The reasoning is: > > 1. Random playouts contains big amount of silly moves > > 2. Hign-dans don't do silly moves, so patterns gathered from > hign-dans games does not contains refutations of such moves > > 3. Most low-dan moves are better then random playout moves > anyway, so patterns gathered from low-dan games are > "good enough" for playouts and may contain refutations > of many low-dan mistakes > > In professional game it is assumed "if opponent make this > suboptimal attack it can be countered with such and such > moves", but this sequence newer happens in the game itself, > so probably good pattern library must be gathered from > games with wide spectrum of player strenghs > ___ > Computer-go mailing list > Computer-go@dvandva.org > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Lars Schaefers Computer Engineering Group of Prof. Dr. Marco Platzner Paderborn Center for Parallel Computing, University of Paderborn Pohlweg 47-49, 33098 Paderborn, Germany Tel: +49 (0)5251 60 4341, Fax: +49 (0)5251 60 5377 Office: Building O 3.119 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Zen!
Nick, thank you for your report and the organization of the tournament! About round 11 you wrote: "In round 11, MCark did not move in its game with gomorra3 until its operator restarted it, after an hour." In fact it was gomorra32 that didn't made its first move because cluster resources expired. I realized it after 2 hours and restarted Gomorra with only 16 cluster nodes. Lars On Sat, 2012-09-15 at 15:53 +, Nick Wedd 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.html and I > look forward, as usual, to receiving your comments and reports of > errors. I apologise for the long delay in uploading this report. > > Nick -- Lars Schaefers Computer Engineering Group of Prof. Dr. Marco Platzner Paderborn Center for Parallel Computing, University of Paderborn Pohlweg 47-49, 33098 Paderborn, Germany Tel: +49 (0)5251 60 4341, Fax: +49 (0)5251 60 5377 Office: Building O 3.119 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] GNU Go command line
Hi Nick, the GnuGo documentation says: "GNU Go can play Monte Carlo Go on a 9x9 board. (Not available for larger boards.)" ( http://www.gnu.org/software/gnugo/gnugo_3.html Section 3.9.2 Monte Carlo Options) Lars On Thu, 2012-06-28 at 12:07 +, Nick Wedd wrote: > I am trying to get GNU Go's monte-carlo option working. > > Running GNU Go 3.8 on KGS, from a Windows system, through khsGtp.jar > if my command line is >engine=gnugo-3.8.exe --level 12 --mode gtp --quiet > it works correctly. But if I add the parameter "--monte-carlo" to get >engine=gnugo-3.8.exe --monte-carlo --level 12 --mode gtp --quiet > it quits with the message >WARNING: Got error response to command "boardsize 19": ? unacceptable > size > > Can anyone explain how to get the --monte-carlo option to work? The > error message is not helpful. > > Nick -- Lars Schaefers Computer Engineering Group of Prof. Dr. Marco Platzner Paderborn Center for Parallel Computing, University of Paderborn Pohlweg 47-49, 33098 Paderborn, Germany Tel: +49 (0)5251 60 4341, Fax: +49 (0)5251 60 5377 Office: Building O 3.119 ___ 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
Aja, thanks for the regression test set. I will make a run with Gomorra. Can you give some details to the sg_compare_float command? What are the parameters, what should it return.. As I couldn't find something in the mail I guess there are no restrictions in terms of time or number of simulations to use. Best wishes, Lars On Tue, 2012-05-15 at 02:39 +, Aja Huang wrote: > Dear all, > > Martin Mueller and I are writing a paper about exploring some > limitations of current MCTS programs in Go. For this purpose we have > carefully designed a regression test set which consists of 20 seki and > 15 two-safe-groups cases on 9x9 board. If you are interested, it is > available at > > http://webdocs.cs.ualberta.ca/~mmueller/ps/seki-and-two-safe-groups-regression-test.zip > > We will appreciate if you would like to run your program over our > regression test and send us the results for our publication. > > It's easy to run your program through these positions (.sgf). Mainly, > the script run.sh under /utility is able to run a given program for a > given regression test file (.tst) and produce the result in a related > html file. For example, for the seki test you can simply type > > ./run.sh -p PATH_TO_PROGRAM -t g_seki_moves.tst > > Some notes: > 1. Your program must support the command sg_compare_float for the > two-safe-groups test. If it doesn't support reg_genmove then the test > file g_seki_moves.tst is good to use which performs genmove instead. > > 2. On Windows platform, you will be able to execute 'run.sh' directly > at the command prompt after cygwin is installed. > > 3. If your program doesn't support the GTP command 'loadsgf', > gogui-adapter is able to translate 'loadsgf' into a sequence of 'play' > commands. The file gogui-adapter.jar under /utility is good to use > because Markus has fixed some bugs for us, see > > https://sourceforge.net/tracker/?func=detail&aid=3522401&group_id=59117&atid=489964 > https://sourceforge.net/tracker/?func=detail&aid=3519829&group_id=59117&atid=489964 > > Under /experimental results, there are results of several programs > such as Fuego (tilburg version), pachi, ManyFaces and GnuGo. We thank > David for providing us the valuable results of ManyFaces. The test set > is really not easy because these programs all failed in many cases. > > Questions are very welcome. If you find any error in the test set > please inform us. Thanks. > > Best regards, > Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Computer Go and EGC 2012
Hi Petr, I'll be most likely there. And of course, a less official tournament and/or an evening chat would be great. So, who else will be there? I'm looking forward to seeing you again! Lars On Wed, 2012-01-18 at 23:03 +, Petr Baudis wrote: > Hi! > > On Wed, Jan 18, 2012 at 08:04:24PM +0100, "Ingo Althöfer" wrote: > > Due to the fact that there will be no special bot tournaments during > > the EGC, there is some more flexibility for the exhibition games. > > Oh, that is quite a disappointment. :-( > > I wonder which program authors will be in Bonn? Maybe we could get > together and organize at least a less official tournament, or just to > chat some evening? Please let us know in this thread if you are coming! > > > For the 19x19 exhibition game the pro will be payed according to handicap. > > He (or she) gets 50 Euro simply for playing. And he has the choice at which > > handicap. When winning the game the pro gets additional money: > > +50 Euro, when handicap was 2, > > +100 Euro, when handicap was 3, > > +200 Euro, when handicap was 4, > > +300 Euro, when handicap was 5. > > That seems kind of strange, what motivation does the pro have to ever > choose a different handicap than 5? > ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Jouni Valkonen's X-Price
Hi Jouni and Ingo, Ingo reminded me of your prize and we, the Gomorra-Team, are actually thinking about it. Gomorra is able to run on compute clusters and recently we made some more progress with our parallelization. It allowed us to come out second in the last two KGS tournaments. For fairness I must state that several strong programs didn't participated in this tournaments like Zen. The KGS account of Gomorra that was also used at the Computer Olympiad is gomorra16. If you want to play it, just send me a mail. In any case, I would say we (and you?) should perhaps wait until about May and see how the programs perform at that time, short before the EGC. Best wishes, Lars On Thu, 2012-01-05 at 18:33 +, "Ingo Althöfer" wrote: > Hello Jouni, > > thanks for your positive reply. > > > Who is your canditate? > > I am not sure if I am allowed to tell now. > I have informed the programmers about your message. > They are a team, so it may take a few days until > they have consensus for a reply. > > > > Also, if particular gobot is new to me, I would like to check it's > > performance at KGS. If it cannot beat me at blitz, it is no use to > > organize slow match either. > > That makes sense. > > One special point with the team is that they hope > to make trememdous progress within the next six months. > An X-price match against you as a reward might just be the > right kick for them to give their very best. > > Ingo. > ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Steenvreter!
Congratulations to Steenvreter! Thank you Nick once again for the organization of the tournament and the report! Gomorra was running on 8 nodes with 12 cores each this time. Sorry Nick, I forgot to give you this information in the registration mail. Also, in the "Details of processor.." section gomorra4 has an "m" too much. This is also the case in the report about the 77th tournament. Lars On Mon, 2011-12-05 at 18:05 +, Nick Wedd wrote: > Congratulations to Steenvreter (playing as 'stv'), winner of yesterday's > 9x9 KGS bot tournament, with 19 wins and one jigo from 20 games! > > My report is at http://www.weddslist.com/kgs/past/78/index.html > As usual I welcome your corrections. > > Nick ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Parallel Monte-Carlo Tree Search for HPC Systems
Hi, for those of you interested in parallel MCTS for distributed memory machines, I want to point you to our paper "Parallel Monte-Carlo Tree Search for HPC Systems" recently presented at the EuroPar 2011 conference. We present a novel approach for the parallelization of MCTS which allows for an equally distributed spreading of both the work and memory load among all compute nodes within a distributed memory HPC systems. This is pretty much the parallelization currently used by our Go Engine Gomorra (KGS account gomorra4). You can download the original paper at Springerlink: http://www.springerlink.com/content/y00906lk33142768/ You can also find an author manuscript on my site: http://www.cs.uni-paderborn.de/fachgebiete/computer-engineering-group/people/schaefers.html Any comments, questions or suggestions are of course welcome. Best wishes, Lars ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to pachi, and to Many Faces of Go!
Hi Nick, Gomorra used a 12 Core SMP machine only. The cluster was used in the last tournament. Thank you for the report! - Lars On Sun, 2011-02-27 at 09:39 +, Nick Wedd wrote: > Congratulations to ManyFaces1, and to pachi2, the joint winners of the > February Slow KGS bot tournament! > > My report is at http://www.weddslist.com/kgs/past/S11.1/index.html > > I hope you will all report the errors that you notice, as usual. > > Nick ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] KGS down
Hi all, for information to those who participate in the KGS SLOW Tournament: Seems that KGS is down since about 1h. It will probably be necessary to restart the engines (if not done in a script anyway). Best wishes, Lars ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Zen!
> BTW Lars, please give me your phone number so that I can call you up when > Gomorra has any trouble. :) I'm a bit scared it may end up in something comparable to telephone terrorism :) - Lars ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Zen!
On Wed, 2011-02-09 at 22:48 -0800, David Doshay wrote: > I agree that any TD should be able to give this kind of advice for > tournaments > like these. Me too. I think we all want to win on the board and not on wrong configuration files of our opponents or comparable things.. I welcome every help that leads to a "normal" tournament play. Best, Lars > > Cheers, > David > > > > On 9, Feb 2011, at 10:15 PM, Darren Cook wrote: > > > Just thought I'd bring this up here, to give others a chance to disagree > > and say if Nick needs to be more hands-off for his tournaments. > > ___ > 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!
Hi Nick, thank you for the report! I only found the old copy-paste error that Gomorra is written Gommorra in the "Details of processor numbers, power, etc" section. Best wishes, Lars On Mon, 2011-02-07 at 23:50 +, Nick Wedd wrote: > Congratulations to Zen19, winner of yesterday's KGS bot tournament! > > My report is at http://www.weddslist.com/kgs/past/68/index.html > I expect it has as many errors as usual, and I look forward to you > reporting them to me. > > Nick ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Monte Carlo (upper confidence bounds appliedto trees)
On Wed, 2010-10-27 at 06:16 +0200, Petr Baudis wrote: > I'm certainly not using progressive unpruning (evaluating only first > f(n) children during move selection). I think it was never used > anywhere > besides Mango and CrazyStone? Gomorra also uses progressive widening/unpruning. But indeed it seems that only few people are using it. For me it works great and helps a lot. Also David Silver's RAVE formular worked best for me. - Lars ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Join 9x9 on CGOS
Wow, I wasn't aware of all the kinds of kaesebrot that exit in the world. But I'm with Terry concerning his believe that the most perfect onces are those that you can get in Brazilian churrascarias. The german one that I meant is the partial sandwich. @Richard: Anyone who supports me with a hint that makes Gomorra at least 50 ELO stronger immediately wins a delicious kaesebrot. For 75 ELO hints you get the chessrole steve mentioned that "kicks ass" and from 100 ELO upwards I'll try to organize a brazilian one for you :) - Lars On Tue, 2010-10-26 at 13:05 -0700, Christoph Birk wrote: > On Tue, 26 Oct 2010, terry mcintyre wrote: > > Cheese-bread, so far as I know, is a bread stuffed with cheese. I often > > enjoy > > them at Brazilian churrascarias. There are many delightful variations. > > 'Kaesebrot' in German is just a slice of break with a slice > of cheese on top (and possibly butter). > The closest translation would be cheese sandwich , except > for the 'sandwich' part (ie. no 2nd layer of bread). > > Christoph > ___ > 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] Join 9x9 on CGOS
On Tue, 2010-10-26 at 11:41 +0200, valky...@phmp.se wrote: > Aha, could it be that Gomorra picked up some good lines from > MyGoFriend and then improves upon them? Because when kaesebrot turned > up I think it really crushed Valkyria for some time. I have now > updated the book used for Valkyria3.5.15-4cx a little bit. For sure it picked some lines, also it seems to me that Gomorra hadn't changed a lot (if at all) in strength by doing that. Probably, next weekend, I'll give it another try. > MyGoFriend plays different from most other programs in the opening, my > guess is that MGF does not suffers from some bias inherent in most MC > programs and thus plays more natural (from human aesthetic point of > view) and stronger moves. perhaps it uses some high quality pattern > matching to bias the tree part and does not use AMAF or RAVE at all. MGF also had a superko-bug in the exp1 Version. It lost at least 2 games against Gomorra with illegal moves. That should be easy to fix if not already done in exp2. > My problem is that AMAF/RAVE makes Valkyria very strong and at the > same time it currently sets a ceiling on how strong it can play > because in many positions it seems to completely miss some very strong > moves systematically and it is very hard to fix that because it so > deeply rooted in an otherwise very efficient selective search process. Same for me and probably everyone working with RAVE :( > > Valkyria3.5.14-4cx is using exactly the same book as in Kanazawa and > Valkyria3.5.15-4cx probably just have a lot of bugfixes and changes to > the playouts. This does not mean that the program now plays stronger. > Sometimes changing the way it behaves in some tactical situations > makes it get into new kinds of fights that reveal more bugs... Also > balancing the ability to judge whether an invasion is possible or not > can be disturbed by adding "better" knowledge to it I believe. At least against Gomorra your modifications were helpful. But as the overall rating of 3.5.15 is below 3.5.14 on CGOS, you should throw your modifications away and give Gomorra a chance again :) Concerning the ability to judge whether an invasion is possible or not, I would expect an enhancement by adding "better" knowledge. From my understanding, a successful invasion most often requires the correct play of a rather long sequence. Here good knowledge is very helpful I think. So, I need an address if you need a cheese-bread :) - Lars > > Best > Magnus > > Quoting Lars Schäfers : > > > Hi Magnus, > > > > kaesebrot is german for cheese-bread and was a test version of Gomorra. > > It has an adaptive opening book and makes a lot of computations after > > every lost game. Thats why it disappeared after every lost game for a > > while. So not a new program. > > Valkyria is constantly doing very well. Also at the olympiade in > > Kanazawa it got a good 5th rank with very limited hardware. I was really > > impressed. > > It appears to me that Gomorra likes the >2600 Elo Versions much more > > than the <2600 Elo Versions of Valkyria. I would give a cheese-bread for > > informations about the difference between Valkyria3.5.14-4cx and > > Valkyria3.5.15-4cx :) > > > > Best wishes, > > Lars > > > > On Tue, 2010-10-26 at 08:48 +0200, valky...@phmp.se wrote: > >> Hi all! > >> > >> The last days there has been some increased activity on CGOS 9x9. > >> MyGoFriend is back and shows it is really strong. I am playing with > >> Valkyria using the opening library developed for the Kanazawa > >> tournament. It is not so helpful yet against MyGoFriend as it plays > >> different line than what I was prepared for. > >> > >> Furthermore other mysterious programs has appeared such as Ghosper and > >> Kaesebrot. At least one of those I think is Mogo given the way it > >> plays the opening. They have not so high rankings yet, maybe because > >> of bugs that makes them lost against very weak programs. They are > >> doing very well against Valkyria and MyGofriend. > >> > >> Right now a strong version of ToGo is playing. So it would be fun to > >> the same kind of activity there was on 19x19 CGOS some time ago. > >> > >> I am also sure the programmers of those unknown programs read this and > >> it would be really nice to have a presentation of the programs here on > >> the list too :-) You do not have to tell all the secrets! > >> > >> Best > >> Magnus > >> ___ > >> Compu
Re: [Computer-go] Join 9x9 on CGOS
Hi Magnus, kaesebrot is german for cheese-bread and was a test version of Gomorra. It has an adaptive opening book and makes a lot of computations after every lost game. Thats why it disappeared after every lost game for a while. So not a new program. Valkyria is constantly doing very well. Also at the olympiade in Kanazawa it got a good 5th rank with very limited hardware. I was really impressed. It appears to me that Gomorra likes the >2600 Elo Versions much more than the <2600 Elo Versions of Valkyria. I would give a cheese-bread for informations about the difference between Valkyria3.5.14-4cx and Valkyria3.5.15-4cx :) Best wishes, Lars On Tue, 2010-10-26 at 08:48 +0200, valky...@phmp.se wrote: > Hi all! > > The last days there has been some increased activity on CGOS 9x9. > MyGoFriend is back and shows it is really strong. I am playing with > Valkyria using the opening library developed for the Kanazawa > tournament. It is not so helpful yet against MyGoFriend as it plays > different line than what I was prepared for. > > Furthermore other mysterious programs has appeared such as Ghosper and > Kaesebrot. At least one of those I think is Mogo given the way it > plays the opening. They have not so high rankings yet, maybe because > of bugs that makes them lost against very weak programs. They are > doing very well against Valkyria and MyGofriend. > > Right now a strong version of ToGo is playing. So it would be fun to > the same kind of activity there was on 19x19 CGOS some time ago. > > I am also sure the programmers of those unknown programs read this and > it would be really nice to have a presentation of the programs here on > the list too :-) You do not have to tell all the secrets! > > Best > Magnus > ___ > 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] Libego
Hi Łukasz > > On Wed, Aug 18, 2010 at 09:23:30AM -0700, Peter Drake wrote: > > Łukasz: > > > > I haven't used Libego directly, but I did take some inspiration from > > it in building the core routines for Orego. > > Same for Pachi. I did not copy or directly transcribe any concrete > code, but libego taught me a lot about writing high-performance go rules > implementation. > Exactly the same for Gomorra. As I told you some time ago, I redesigned the core routines after having a look at libego's source. However, nothing was copied or transcribed directly. Cheers, Lars ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] name for distance metric
Arthur, you probably saw it in this paper of Remi Coulom: http://remi.coulom.free.fr/Amsterdam2007/MMGoPatterns.pdf It is used to define diamond shaped patterns there. Best wishes, Lars On Fri, 2010-06-18 at 09:28 +0100, Arthur Cater wrote: > Is there a formal name for this distance metric? > > distance=C.(dx+dy+max(dx,dy)) > > where C may be 1 or 1/2 , it doesn't much matter. > > And in which well-known Go program(s) is it used? I'm sure I've seen > something > about it recently but damned if I can find it again. > > Help much appreciated. > > Arthur > ___ > 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 Zen9!
Nick, There are copy-paste errors in the "Details of processor numbers.." section: - A typo in gomorra9 (gommorra9). - Gomorra ran with 8 threads on two Core i7 CPUs (not using hyperthreading). Thank you for organizing the tournament!! - Lars On Mon, 2010-05-03 at 20:12 +0100, Nick Wedd wrote: > Congratulations to Zen9, winner of yesterday's KGS bot tournament with > 16 wins from 18 games! > > I found this an exciting tournament. My report, somewhat longer than > usual, is now at http://www.weddslist.com/kgs/past/59/index.html > > I look forward to your reports of my errors, as usual. > > Nick ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go