Re: [Computer-go] MCTS playouts per second
On 19x19, Erica's speed is around 5,500 lightweight playouts per second on a single i7 cpu. As far as I know, Lukasz Lew's libego, which is open source, is the fastest implementation of MCTS and can reach around 6,000-7,000 lightweight playouts per second in the same cpu. Aja -原始郵件- From: Scott Christensen Sent: Wednesday, October 26, 2011 6:48 AM To: computer-go@dvandva.org Subject: [Computer-go] MCTS playouts per second Just want to check what the expected playout performance is of well tuned monte-carlo engines? My MCTS engine is averaging apx 3,500 lightweight playouts per second on a single i5 32 bit cpu. Any suggestions on very efficient source code examples for fast monte-carlo playouts? I've spent a lot of time comparing recursive group formation vs non-recursive but it doesn't seem to make a big difference. It seems that updating the list of likely moves after every play with something similar to the mogo probability rules is the most time consuming part as I currently recalculate the probabilities of moves at every empty point on the board each turn. It seems necessary if one doesn't want to handle all the exceptions to keeping the previous turn's play probabilities. Also any thoughts on combining pattern scoring and other conventional techniques together with a UCT tree? If two branches have very similar simulated win ratios could one use other factors to choose the best branch? It seems if there is a very wide branching such as at the beginning of the game, there is a lot of room to add other heuristics to choosing the best move when monte-carlo scores are within range of expected error. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] MCTS playouts per second
No, I meant 6,000-7,000 playouts per second on 19x19. Aja From: Michael Williams Sent: Thursday, October 27, 2011 5:18 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] MCTS playouts per second Perhaps you meant to say 60,000-70,000 playouts per second for libego? On Wed, Oct 26, 2011 at 10:23 PM, Aja Huang ajahu...@gmail.com wrote: On 19x19, Erica's speed is around 5,500 lightweight playouts per second on a single i7 cpu. As far as I know, Lukasz Lew's libego, which is open source, is the fastest implementation of MCTS and can reach around 6,000-7,000 lightweight playouts per second in the same cpu. Aja -原始郵件- From: Scott Christensen Sent: Wednesday, October 26, 2011 6:48 AM To: computer-go@dvandva.org Subject: [Computer-go] MCTS playouts per second Just want to check what the expected playout performance is of well tuned monte-carlo engines? My MCTS engine is averaging apx 3,500 lightweight playouts per second on a single i5 32 bit cpu. Any suggestions on very efficient source code examples for fast monte-carlo playouts? I've spent a lot of time comparing recursive group formation vs non-recursive but it doesn't seem to make a big difference. It seems that updating the list of likely moves after every play with something similar to the mogo probability rules is the most time consuming part as I currently recalculate the probabilities of moves at every empty point on the board each turn. It seems necessary if one doesn't want to handle all the exceptions to keeping the previous turn's play probabilities. Also any thoughts on combining pattern scoring and other conventional techniques together with a UCT tree? If two branches have very similar simulated win ratios could one use other factors to choose the best branch? It seems if there is a very wide branching such as at the beginning of the game, there is a lot of room to add other heuristics to choosing the best move when monte-carlo scores are within range of expected error. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] MCTS playouts per second
Thanks Rémi's clarification. I forgot that 6,000-7,000 playouts per second is with supporting weighted moves globally and incrementally updating various data structures, not pure uniform random playouts. If I recall correctly, Lukasz's implementation was 1,000-2,000 playouts per second faster than mine. Aja -原始郵件- From: Rémi Coulom Sent: Friday, October 28, 2011 2:33 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] MCTS playouts per second I think it is more like 100k on 9x9, and 25k on 19x19 http://www.mail-archive.com/computer-go@computer-go.org/msg11214.html Rémi On 28 oct. 2011, at 06:30, Aja Huang wrote: No, I meant 6,000-7,000 playouts per second on 19x19. Aja From: Michael Williams Sent: Thursday, October 27, 2011 5:18 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] MCTS playouts per second Perhaps you meant to say 60,000-70,000 playouts per second for libego? On Wed, Oct 26, 2011 at 10:23 PM, Aja Huang ajahu...@gmail.com wrote: On 19x19, Erica's speed is around 5,500 lightweight playouts per second on a single i7 cpu. As far as I know, Lukasz Lew's libego, which is open source, is the fastest implementation of MCTS and can reach around 6,000-7,000 lightweight playouts per second in the same cpu. Aja -原始郵件- From: Scott Christensen Sent: Wednesday, October 26, 2011 6:48 AM To: computer-go@dvandva.org Subject: [Computer-go] MCTS playouts per second Just want to check what the expected playout performance is of well tuned monte-carlo engines? My MCTS engine is averaging apx 3,500 lightweight playouts per second on a single i5 32 bit cpu. Any suggestions on very efficient source code examples for fast monte-carlo playouts? I've spent a lot of time comparing recursive group formation vs non-recursive but it doesn't seem to make a big difference. It seems that updating the list of likely moves after every play with something similar to the mogo probability rules is the most time consuming part as I currently recalculate the probabilities of moves at every empty point on the board each turn. It seems necessary if one doesn't want to handle all the exceptions to keeping the previous turn's play probabilities. Also any thoughts on combining pattern scoring and other conventional techniques together with a UCT tree? If two branches have very similar simulated win ratios could one use other factors to choose the best branch? It seems if there is a very wide branching such as at the beginning of the game, there is a lot of room to add other heuristics to choosing the best move when monte-carlo scores are within range of expected error. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Elastic Cloud
In the UEC Cup last weekend, I rented a machine from Amazon EC2 to run Erica: High-CPU Extra Large Windows Instance (c1.xlarge) $1.16 per/hour (High-CPU Extra Large Instance 7 GB of memory, 20 EC2 Compute Units (8 virtual cores with 2.5 EC2 Compute Units each), 1690 GB of local instance storage, 64-bit platform) However, according to my testing, this machine is around 1.5 times slower than my 4-core (i7 950) PC. It is a high-level instance aleady but actually very slow. A competition on equal hardware from Amazon EC2 might be interesting, but such weak hardware might not reflect program's performance on large simulations. Aja -原始郵件- From: Mark Boon Sent: Wednesday, December 07, 2011 11:13 PM To: computer-go@dvandva.org Subject: [Computer-go] Elastic Cloud I haven't had time to concern myself with computer-Go the past years. I even have a hard time keeping up with this list. But today I had a few hours to kill. For work I deal a lot with the Amazon EC2 service these days. And I noticed they're not billing us for the micro instances. We mostly use their large instances anyway. It turns out for the first year Amazon doesn't charge for a micro instance. You probably need a credit card to open an account but then you have a free Linux instance for a year. Normally they charge $0.02 per hour (comes down to $15 for a month full-time use) or less when you 'reserve' usage ahead of time. I remember a while back there was some discussion about holding a competition on equal hardware. This may be a good and cheap, maybe even free, way to arrange something like this. I didn't really have a clue as to how fast these micro instances are so while I had a little time I decided to give it a little spin. I set up a micro instance and installed some basic packages on it, like git and make. I downloaded Libego, compiled it (something I never managed on my Mac) and ran its benchmark. It literally didn't take me more than 15 minutes in all from starting the instance to running Libego. It reported 27kpps and 10kpps per Ghz. This seems to point to a 2.66Ghz processor underneath. Not terribly exciting, but not too bad either (hey, it's free!). The numbers seemed a bit low. So I looked in the code and modified it so that it did just playouts and didn't use the sampler (I didn't bother to find out what the sampler was for) and the numbers improved a bit to 44kpps and 16kkps per Ghz. Next I uploaded a Java jar with the refbot I made some years back. That recorded 27kkps for plain playouts. It does compute real liberties, not pseudo like Libego, but that's minor. My Mac does around 11-12 kkps per Ghz for the Java version, so that looked fairly normal to me. I thought Libego did more like 40 kkps per Ghz, so that surprised me a bit. Maybe others can tell me if this sounds about right for Libego. Anyway, I thought maybe some if you might be interested in this. I know that not everyone would favor a competition on equal hardware, but I think it would be an interesting alternative. Mark Boon ___ 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] Elastic Cloud
By the way, they charge $0.02/hour for running a Micro Windows Instance. Aja___ 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 ofsims in 9x9 (unlimited time setting for humans)
I just gave a quick try. The current version of Erica scores 75.4%(±3.2) against GnuGo 3.8 Level 10 on 9x9, with 300 playouts/move. It can be even stronger since the parameters are well tuned for 19x19 but not 9x9. I think Zen and Crazy Stone can reach higher winning rate. Aja From: Brian Sheppard Sent: Saturday, December 10, 2011 2:47 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] level of MCTS programs depending on the number ofsims in 9x9 (unlimited time setting for humans) Pebbles is in between Mogo and Fuego in terms of simulation count and CPU effort. E.g., Pebbles-800 is better than Mogo-1000 and worse than Fuego-1000. My impression is that Erica needs even fewer trials. From the Erica papers I recall that Erica-300 was a formidable player. E.g., better than GnuGo (1800-rated). It is clear that Valkyria also needs very few trials to hit high ratings on 9x9. Brian From: computer-go-boun...@dvandva.org [mailto:computer-go-boun...@dvandva.org] On Behalf Of Olivier Teytaud Sent: Saturday, December 10, 2011 8:19 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] level of MCTS programs depending on the number of sims in 9x9 (unlimited time setting for humans) 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,p2657 Fuego0.4.1-30K,p2520 Fuego0.4.1-10K,p2376 Fuego0.4.1-3K,p2205 Fuego0.4.1-1K,p2024 Fuego0.4.1-300,p1756 Mogo3MC90K,p2428 Mogo3MC30K,p2304 Mogo3MC10K,p2169 Mogo3MC3K,p 2008 Mogo3MC1K,p1823 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] CrazyStone in the 5-dan footsteps of Zen
I have played several games with the 5 dan CrazyStone (24 cores) and felt that it is pretty close to the 5 dan Zen (26 cores) in playing strength. In fact, CrazyStone’s playing style is much more balanced and human-like than Zen. Specifically, CrazyStone is better than Zen at winning without killing or fighting. It has a very good sense of territory and features very good pattern shapes, though it is still weaker than Zen in handling semeais. Blitz games such as 15s/move favor MCTS programs. I expect both Zen and CrazyStone will drop to 4d in longer games. The reason is that MCTS programs are still not able to (in my opinion) read deepily like human players in situations such as big semeai or life-and-death. Also, when against strong Go players in fast games MCTS programs rely heavily on accurate territory counting. In longer games, this advantage will decrease most if not at all. Aja From: Don Dailey Sent: Monday, January 02, 2012 3:28 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen On Mon, Jan 2, 2012 at 4:02 PM, Ingo Althöfer 3-hirn-ver...@gmx.de wrote: Hello Don, What time control was used for these games or did it vary? In the last few days it was 15 seconds per move, with 9 byoyomi phases. I don't think that specifies the time control does it?So if a player exceeds 15 seconds he starts to use one of his byoyomi periods?How long for the byoyomi phases? I don't think 15 seconds per move average is bad for humans at all, but if you have to make each move in 15 seconds it's horrible and I can see that this is going to make the computer really look good. For making strength claims there should be some sort of standard. You could take almost any program and get 5 dan or more by setting the time to 1 second per move. Don Ingo. -- NEU: FreePhone - 0ct/min Handyspartarif mit Geld-zurück-Garantie! Jetzt informieren: http://www.gmx.net/de/go/freephone ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] replacing dynamic komi with a scoring function
No long ago I was considering a scheme something like bias uct formula by solid points to cure the problem of underestimation of the edge and corner territory, but still have no time to try. I believe it is one of the solutions to the notorious problem that MCTS programs usually like center territory more. Take the empty position as an example: a center move A and a corner move B might both have 47% winning rate, but B should have higher score in UCT formula computation. The reason is simple: from the ownership information of the playouts, there is no solid points for A (everywhere both sides own almost 50%) while B has some solid points at the corner (such as Black owns that corner points in 60% of the playouts). Aja -原始郵件- From: Yamato Sent: Sunday, January 08, 2012 8:30 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] replacing dynamic komi with a scoring function (2012/01/09 9:56), Jouni Valkonen wrote: There is indeed a problem with Dynamic komi with Zen. Zen often loses the handicap games if black tries to minimize the move count. Often if it is possible to bring game to small yose in around move 180 or so and if not too much behind, then Zen most likely will lose. I have played few games where Zen noted only when filling last dames that it is losing the game by few or half points and then resign. Although, one game was that I lost by ½ points, because I accidentally defended unnecessarily instead of taking the last dame. One game was that i was about ten points behind around move 180, but then Zen played a slack small yose, and lost by 2½ points. Also good and very easy strategy against zen in handicap games is to take all the sides and give center territory to the Zen. Zen almost always will take the center territory as too small and gives sides as too big. I think people often confuse the evaluation problem and the dynamic komi problem. A more urgent problem is the underestimation of the edge and corner territory. -- Yamato ___ 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] replacing dynamic komi with a scoring function
Yes, computer try the win the game, but right now MCTS have problems by simply using winning rate in 19x19 Go. That’s why we are using “dynamic kom”, which is collected/computed from average score”, to cure this problem. My point is that using average score is maybe not enough. We might also have to use the information of “location” from the “ownership map” (looks to me a nice word). Aja From: Don Dailey Sent: Monday, January 09, 2012 11:24 AM To: Aja Huang ; computer-go@dvandva.org Subject: Re: [Computer-go] replacing dynamic komi with a scoring function On Mon, Jan 9, 2012 at 12:54 PM, Aja Huang ajahu...@gmail.com wrote: It’s easy to detect “secure” territory by collecting the information of ownership from the playouts. I have always called this the ownership map for lack of a better term. I have heard people refer to is as a futures map. If Black owns a region of points in 95% of the playouts, for example, we can “safely” say this region is Black’s territory (unless the playouts are seriously biased, which is independent to this problem). Go is a game of territory and the objective of every move is to gain more territory. Computers have a different objective.Computer try to win the game, humans trying to win territory. To cure this problem, besides winning rate we might have to use the information of not only “average score” but also “average score of certain points”. I fear the program would find ways to trade some territory in for other. It does seem like there should be some way to use this information to help. Don Aja From: Stefan Kaitschick Sent: Monday, January 09, 2012 4:38 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] replacing dynamic komi with a scoring function By the way, are we sure it is underestimation of the edges and corners ? Rather than overestimation of the centre ? I know those are equivalent for play itself, but the answer suggests different tries for solution. In the first case, we want to make the bot more aware that he will keep its edge territory. In the second, we want to make it understand that inways made be made in its beautiful centre. Jonas I think they overestimate both the corners and the center, but they overestimate the corners less. :-) They will often force from the outside, even when an invasion is relatively simple and the outside forcing stones aren't worth much. I can only call that overestimating corner safety. But at the same time, the center is given even greater priority, because the playouts so often come back with a kill of would be invaders. Crazy Stone seems to be ahead of the other bots in this. My guess is that it's using simulation balancing in the playouts to purposely degrade attacking moves. Ajas' idea of biasing towards secure territory is a typical strong players idea. But what does secure mean? Maybe the idea Ingo brought up, to naively give corner and edge territory a higher weight in the early stages of the game, is more promising. It feels like a crutch, but when you have broken leg, a crutch is great. Stefan -- ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] replacing dynamic komi with a scoring function
Interesting. I have observed a problem in the current dynamic komi scheme especially when there is a big semeai/life-and-death: Playout 1 B+0.5 Playout 2 B+0.5 Playout 3 B+0.5 Playout 4 B+0.5 Playout 5 B+0.5 Playout 6 B+0.5 Playout 7 B+0.5 Playout 8 B+0.5 Playout 9 B+0.5 Playout 10 W+8.5 B has 90% winning rate. How many points should we shift for dynamic komi? Aja -原始郵件- From: Zach Wegner Sent: Monday, January 09, 2012 11:53 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] replacing dynamic komi with a scoring function On Mon, Jan 9, 2012 at 7:26 AM, Vlad Dumitrescu vladd...@gmail.com wrote: Hi, On Mon, Jan 9, 2012 at 13:17, Don Dailey dailey@gmail.com wrote: Summary: I believe a more correct scoring function won't be based on how much you win by OR how often you win but will incorporate some other more relevant concept and it will be dynamic.And it will not matter if the game is a handicap game or otherwise because the scoring function will always be relevant. The goal will be to maximize your winning chances but it will incorporate something more sophisticated that just counting how often you win or how much you win by. I hope I may interfere with something that Don's nice description revealed to me. It feels rather obvious, but since nobody stated it explicitly, maybe it's news for at least some people here. MCTS is maximizing the chances of winning. These chances are largest for a minimal score difference because this allows for making some errors. Winning by the largest possible score has rather small chances to happen because every move has to be perfect. The curve describing the probability of ending the game with a certain score is bell-shaped and MCTS explores the area beneath it, looking for winning moves. With handicap, the disadvantaged side is getting less samples explored, making it less likely to discover the really good moves. Dynamic komi shifts the bell left or right in order to equalize the sampling on both sides, but as mentioned it isn't dynamic enough (the curve changes after each move) and also is actually using a different shape for the curve than the real handicap curve. In theory, I think that the solution for keeping the same level of play with handicap as without would be to make sure that the the disadvantaged side gets just as many samples with or without handicap. That is, use more playouts when playing with handicap. In practice, this is probably prohibitive... I wonder if it might be possible to estimate the shape of this curve after each move and use that estimate to dynamically adjust the number of playouts. One might have to use higher precision calculations, too, so that the noise doesn't get too loud. Does this make any sense? Has anyone tried something like this? best regards, Vlad ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go This is similar to ideas I've had to solve this. Of course, I'm not a go programmer, so it's really just idle curiosity for me when reading this list. Perhaps an ideal scoring function is highest median score. You can do this in a slow/memory intensive way by storing a histogram of final score in each node instead of wins/losses. Wins/losses can then be computed by summing the histogram buckets above/below the komi level. Computing the median is simply finding the bucket where the sum crosses 50%. Other percentages could be used here as well, though I'm not quite sure what effect this might have. This solves the problem of getting far less than one bit of information per playout when the win rate is very high or low, and also supersedes dynamic komi--if you have the full histogram at any node, you can compute the exact win rate for any arbitrary komi. Using this in practice might be fairly difficult, and would probably involve making the histogram lossy in some sense. I would be interested if somebody here could run tests with this as a limit study though--keep the number of nodes allocated and playouts per move constant, and see the effect on strength. Zach ___ 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] replacing dynamic komi with a scoring function
Lets say there is a semeai and Black wins 6.5 points if it wins the semeai. Otherwise White wins 3.5 points. Playout 1B+6.5 Playout 2B+6.5 Playout 3B+6.5 Playout 4B+6.5 Playout 5B+6.5 Playout 6B+6.5 Playout 7B+6.5 Playout 8B+6.5 Playout 9B+6.5 Playout 10 W+3.5 Average score = (6.5 * 9 + (- 3.5)) / 10 = 5.5 In this case, shifting komi to something like “Black 0.5 point win” (namely increasing komi by 5) does not work. Black simply has to win this semeai to win. So I think Zach’s idea quite makes sense. Actually I had a similar idea in the past but still have no time to try it. Aja From: Stefan Kaitschick Sent: Monday, January 09, 2012 2:28 PM To: Aja Huang ; computer-go@dvandva.org Subject: Re: [Computer-go] replacing dynamic komi with a scoring function The correct answer is 0, because shifting by just 1 point drops the rate to 10% But is this really a dynamic komi problem? I mean, is there really a graceful way to misevaluate semeais? apropos semeais: would it be feasible to expand RAVE to include the information of the previous move? ( like the killer heuristic in chess) It could be limited to the 5 by 5 surroundings, so that the previous move is one of 24 points or other. But that's still not cheap computationally. I just have no idea how much effort it takes to maintain the current RAVE. Stefan On Mon, Jan 9, 2012 at 8:14 PM, Aja Huang ajahu...@gmail.com wrote: Interesting. I have observed a problem in the current dynamic komi scheme especially when there is a big semeai/life-and-death: Playout 1 B+0.5 Playout 2 B+0.5 Playout 3 B+0.5 Playout 4 B+0.5 Playout 5 B+0.5 Playout 6 B+0.5 Playout 7 B+0.5 Playout 8 B+0.5 Playout 9 B+0.5 Playout 10 W+8.5 B has 90% winning rate. How many points should we shift for dynamic komi? Aja -原始郵件- From: Zach Wegner Sent: Monday, January 09, 2012 11:53 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] replacing dynamic komi with a scoring function On Mon, Jan 9, 2012 at 7:26 AM, Vlad Dumitrescu vladd...@gmail.com wrote: Hi, On Mon, Jan 9, 2012 at 13:17, Don Dailey dailey@gmail.com wrote: Summary: I believe a more correct scoring function won't be based on how much you win by OR how often you win but will incorporate some other more relevant concept and it will be dynamic.And it will not matter if the game is a handicap game or otherwise because the scoring function will always be relevant. The goal will be to maximize your winning chances but it will incorporate something more sophisticated that just counting how often you win or how much you win by. I hope I may interfere with something that Don's nice description revealed to me. It feels rather obvious, but since nobody stated it explicitly, maybe it's news for at least some people here. MCTS is maximizing the chances of winning. These chances are largest for a minimal score difference because this allows for making some errors. Winning by the largest possible score has rather small chances to happen because every move has to be perfect. The curve describing the probability of ending the game with a certain score is bell-shaped and MCTS explores the area beneath it, looking for winning moves. With handicap, the disadvantaged side is getting less samples explored, making it less likely to discover the really good moves. Dynamic komi shifts the bell left or right in order to equalize the sampling on both sides, but as mentioned it isn't dynamic enough (the curve changes after each move) and also is actually using a different shape for the curve than the real handicap curve. In theory, I think that the solution for keeping the same level of play with handicap as without would be to make sure that the the disadvantaged side gets just as many samples with or without handicap. That is, use more playouts when playing with handicap. In practice, this is probably prohibitive... I wonder if it might be possible to estimate the shape of this curve after each move and use that estimate to dynamically adjust the number of playouts. One might have to use higher precision calculations, too, so that the noise doesn't get too loud. Does this make any sense? Has anyone tried something like this? best regards, Vlad ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go This is similar to ideas I've had to solve this. Of course, I'm not a go programmer, so it's really just idle curiosity for me when reading this list. Perhaps an ideal scoring function is highest median score. You can do this in a slow/memory intensive way by storing a histogram of final score in each node instead of wins/losses
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
When a program becomes famous/interesting on KGS, some users would try to play with it. I have seen that one guy registered a new account and used that account to beat Zen from 6 stones until 2 stones. Now sure if it hurts Zen's rating much. Just now I saw an account StoneCrazy [4d] on KGS From the view of a Go programmer, I would expect that KGS can support reject a match for a ranked robot. Aja -原始郵件- From: Rémi Coulom Sent: Tuesday, January 10, 2012 2:43 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen I wonder what kind of abuse you censor. I never observed any behaviour I'd like to censor with Crazy Stone. Rémi On 10 janv. 2012, at 21:19, Jean-loup Gailly wrote: I know the Zen author occasionally logs in as the bot and censors people who abuse the bot. I do this too for pachi2. But it's a real pain to maintain the censor list. My list currently has 147 accounts (probably representing much fewer distinct persons). Jean-loup ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen
Hi Vlad, Do you mean the case in the attached example semeai_scoring.sgf? In a seki of a dead group with a square of four, the other side must have a big eye as well. The bottom-right corner is such an example, where it's all White's territory. In the bottom-left corner. It can't be a seki. Best regards, Aja -原始郵件- From: Vlad Dumitrescu Sent: Wednesday, January 11, 2012 10:11 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] CrazyStone in the 5-dan footsteps of Zen Hi Rémi, On Wed, Jan 11, 2012 at 16:28, Rémi Coulom remi.cou...@free.fr wrote: Sorry for your medal. I don't do the UI stuff, but I'll forward your remark to Unbalance. Or if you can send the position with the wrong scoring to me, I'll try to make CS score it correctly. It's not a problem for me, but thought it would be a simple improvement. I don't have the position, but it was a semeai where CS's dead group had a square of four in the corner but not enough liberties overall to make a seki. And since the counting seems to be Japanese (since prisoners are deducted), I would have lost points by capturing the group effectively. regards, Vlad ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go semeai_scoring.sgf Description: application/go-sgf ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Zen blunder in long mirror go game
In the last UEC Cup, the Japanese program katsunari played with mirror Go strategy against Zen, Fuego and ManyFaces. katsunari lost all three games. http://jsb.cs.uec.ac.jp/~igo/eng/result_2nd/Zen-katsunari.sgf.html http://jsb.cs.uec.ac.jp/~igo/eng/result_2nd/katsunari-Fuego.sgf.html http://jsb.cs.uec.ac.jp/~igo/eng/result_2nd/katsunari-TheManyFacesofGo.sgf.html Note that Fuego lost the game finally (by filling own territories) because I carelessly set the config with Chinese rules. Anyway, there should be no problem for current strong Go programs on mirror Go strategy in EVEN or tournament games. Aja -原始郵件- From: Yamato Sent: Monday, January 16, 2012 5:16 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Zen blunder in long mirror go game (2012/01/17 8:43), Michael Williams wrote: If the author looks at it, could you update us? I'm always curious about how such things happen, especially in a bot as strong as Zen. Honestly I don't want to spend a lot of time on this type of problem. -- Yamato ___ 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 blunder in long mirror go game
I am not talking about that Zen’s “blunder” but simply mentioned in passing about handling mirroring strategy that Yamato and we don’t want to spend time with. For this game, you should be strong enough to judge Zen was winning or not. Aja From: Jouni Valkonen Sent: Monday, January 16, 2012 5:33 PM To: Aja Huang ; computer-go@dvandva.org Subject: Re: [Computer-go] Zen blunder in long mirror go game I think that the main problem was not winning or losing a mirrored game but the blunder in late endgame by playing suicidal move. Perhaps Zen was losing the game, so this was the reason for the blunder. But if Zen was winning, then it was serious bug and it had nothing to do with the mirroring itself. —Jouni On Jan 17, 2012 2:26 AM, Aja Huang ajahu...@gmail.com wrote: In the last UEC Cup, the Japanese program katsunari played with mirror Go strategy against Zen, Fuego and ManyFaces. katsunari lost all three games. http://jsb.cs.uec.ac.jp/~igo/eng/result_2nd/Zen-katsunari.sgf.html http://jsb.cs.uec.ac.jp/~igo/eng/result_2nd/katsunari-Fuego.sgf.html http://jsb.cs.uec.ac.jp/~igo/eng/result_2nd/katsunari-TheManyFacesofGo.sgf.html Note that Fuego lost the game finally (by filling own territories) because I carelessly set the config with Chinese rules. Anyway, there should be no problem for current strong Go programs on mirror Go strategy in EVEN or tournament games. Aja -原始郵件- From: Yamato Sent: Monday, January 16, 2012 5:16 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Zen blunder in long mirror go game (2012/01/17 8:43), Michael Williams wrote: If the author looks at it, could you update us? I'm always curious about how such things happen, especially in a bot as strong as Zen. Honestly I don't want to spend a lot of time on this type of problem. -- Yamato ___ 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 blunder in long mirror go game
OK, I see your point. So it should be Yamato reply the question if he like. :) Aja From: Jouni Valkonen Sent: Monday, January 16, 2012 6:45 PM To: Aja Huang ; computer-go Subject: Re: [Computer-go] Zen blunder in long mirror go game Aja, you are talking something that has zero relevance for this topic. Zen was winning by ½ point, but it miscalculated in late yose and blundered severely. This was serious endgame bug. This thread has absolutely nothing to do with the fact that the game was mirrored, but only that that Zen throw away the game in late yose. I have seen in my own games several times, when Zen has blundered in late yose and have lost a won games. I think that it is due that zen is winning by half point or so, but miscalculates and then tries something irrational. It would be good idea to analyze, to see what went wrong with Zen's evaluation of the score. Was that it miscalculated the territory or did it misread the seki thing? –Jouni On 17 January 2012 02:44, Aja Huang ajahu...@gmail.com wrote: I am not talking about that Zen’s “blunder” but simply mentioned in passing about handling mirroring strategy that Yamato and we don’t want to spend time with. For this game, you should be strong enough to judge Zen was winning or not. Aja From: Jouni Valkonen Sent: Monday, January 16, 2012 5:33 PM To: Aja Huang ; computer-go@dvandva.org Subject: Re: [Computer-go] Zen blunder in long mirror go game I think that the main problem was not winning or losing a mirrored game but the blunder in late endgame by playing suicidal move. Perhaps Zen was losing the game, so this was the reason for the blunder. But if Zen was winning, then it was serious bug and it had nothing to do with the mirroring itself. —Jouni On Jan 17, 2012 2:26 AM, Aja Huang ajahu...@gmail.com wrote: In the last UEC Cup, the Japanese program katsunari played with mirror Go strategy against Zen, Fuego and ManyFaces. katsunari lost all three games. http://jsb.cs.uec.ac.jp/~igo/eng/result_2nd/Zen-katsunari.sgf.html http://jsb.cs.uec.ac.jp/~igo/eng/result_2nd/katsunari-Fuego.sgf.html http://jsb.cs.uec.ac.jp/~igo/eng/result_2nd/katsunari-TheManyFacesofGo.sgf.html Note that Fuego lost the game finally (by filling own territories) because I carelessly set the config with Chinese rules. Anyway, there should be no problem for current strong Go programs on mirror Go strategy in EVEN or tournament games. Aja -原始郵件- From: Yamato Sent: Monday, January 16, 2012 5:16 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Zen blunder in long mirror go game (2012/01/17 8:43), Michael Williams wrote: If the author looks at it, could you update us? I'm always curious about how such things happen, especially in a bot as strong as Zen. Honestly I don't want to spend a lot of time on this type of problem. -- Yamato ___ 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] KGS, gogui and handicap stones
A simple solution would be to keep a variable HandicapStone.
In the function of GTP command play:
If (WhiteNeverPlayed == true) // WhiteNeverPlayed is initialized to true
{
if (Color == BLACK
Move != PASS)
HandicapStone += 1.0f; // HandicapStone is initialized to 0
else
WhiteNeverPlayed = false;
}
Then in the scoring function:
if (HandicapStone = 2)
BlackScore -= (HandicapStone / 2);
Aja
-原始郵件-
From: David Fotland
Sent: Saturday, January 21, 2012 11:36 AM
To: computer-go@dvandva.org
Subject: Re: [Computer-go] KGS, gogui and handicap stones
I have code that detects several black moves in a row at the start of the
game, and sets the proper handicap property for the engine.
David
-Original Message-
From: computer-go-boun...@dvandva.org [mailto:computer-go-
boun...@dvandva.org] On Behalf Of ds
Sent: Saturday, January 21, 2012 2:26 AM
To: computer-go@dvandva.org
Subject: [Computer-go] KGS, gogui and handicap stones
Hi,
sounds simple, but I could not figure out:
If I load a game from kgs with gogui, my engine is not informed about
the handicap stones. Therefore it counts wrong? Is there a workaround?
Thanks
Detlef
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] KGS, gogui and handicap stones
A player might pass when putting the handicap stones. So here is the correct
code:
In the function of GTP command play:
if (WhiteNeverPlayed == true) // WhiteNeverPlayed is initialized to
true
{
if (Color == BLACK)
{
if (Move != PASS)
HandicapStone += 1.0f; // HandicapStone is initialized to
0
}
else
WhiteNeverPlayed = false;
}
Then in the scoring function:
if (HandicapStone = 2)
BlackScore -= (HandicapStone / 2);
Just call play in the GTP commands for handicap setting.
Aja
-原始郵件-
From: Aja Huang
Sent: Saturday, January 21, 2012 12:54 PM
To: computer-go@dvandva.org
Subject: Re: [Computer-go] KGS, gogui and handicap stones
A simple solution would be to keep a variable HandicapStone.
In the function of GTP command play:
If (WhiteNeverPlayed == true) // WhiteNeverPlayed is initialized to true
{
if (Color == BLACK
Move != PASS)
HandicapStone += 1.0f; // HandicapStone is initialized to 0
else
WhiteNeverPlayed = false;
}
Then in the scoring function:
if (HandicapStone = 2)
BlackScore -= (HandicapStone / 2);
Aja
-原始郵件-
From: David Fotland
Sent: Saturday, January 21, 2012 11:36 AM
To: computer-go@dvandva.org
Subject: Re: [Computer-go] KGS, gogui and handicap stones
I have code that detects several black moves in a row at the start of the
game, and sets the proper handicap property for the engine.
David
-Original Message-
From: computer-go-boun...@dvandva.org [mailto:computer-go-
boun...@dvandva.org] On Behalf Of ds
Sent: Saturday, January 21, 2012 2:26 AM
To: computer-go@dvandva.org
Subject: [Computer-go] KGS, gogui and handicap stones
Hi,
sounds simple, but I could not figure out:
If I load a game from kgs with gogui, my engine is not informed about
the handicap stones. Therefore it counts wrong? Is there a workaround?
Thanks
Detlef
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Beating old ManyFaces at 29 handicap stones
You should let me try. :) Aja From: David Fotland Sent: Saturday, January 28, 2012 6:45 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones I tried 29 stones once and it crushed me J David From: computer-go-boun...@dvandva.org [mailto:computer-go-boun...@dvandva.org] On Behalf Of Michael Williams Sent: Saturday, January 28, 2012 5:43 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones Nice. This should be interesting... On Sat, Jan 28, 2012 at 8:33 PM, David Fotland fotl...@smart-games.com wrote: mfgo1998 is up on kgs if anyone wants to try it. To get a big handicap, set up a 9 stone game and pass until you get up to 29 stones in the same pattern as Martin's game. Regards, David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of David Fotland Sent: Saturday, January 28, 2012 11:34 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones The game with Martin was played on 8/5/1998 at the US Go Congress. My closest source is dated 7/3/1998. I have this built into a gtp front end now, and it's playing different moves from Martin's game, so the engine is not identical. Do you still want to continue with this bet? David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Ingo Althöfer Sent: Thursday, January 26, 2012 1:22 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones Which old Manyface, I have a DOS version from 1990 :) In 1997 ManyFaces 10.0 came to the market (and is sol still nowadays). David Fotland wrote that in the game against Martin Mueller he likely used a source from July 1998. He will try to reanimate that. ingo. -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ 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] Beating old ManyFaces at 29 handicap stones
Yes, thanks. Actually in the last try, I lost only 30 points and was having a chance as well. ManyFaces is not unbeatable with 29 stones. I also tried 2 games against Minirock2 with 6 stones, 1 win 1 loss. The first game was a test. Minirock2 is obviously a MCTS program and it handles ladders very well. In the second, I killed a big group. Minirock2 also has the popular semeai problem. I think its playing strength is maybe around 1d. My guess would be Minirock2 is the Japanese program blast. Aja From: Stefan Kaitschick Sent: Saturday, January 28, 2012 8:41 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones Awesome attempt! 285 at R8 and you would have crushed it in the first try. Stefan On Sun, Jan 29, 2012 at 2:47 AM, Aja Huang ajahu...@gmail.com wrote: You should let me try. :) Aja From: David Fotland Sent: Saturday, January 28, 2012 6:45 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones I tried 29 stones once and it crushed me J David From: computer-go-boun...@dvandva.org [mailto:computer-go-boun...@dvandva.org] On Behalf Of Michael Williams Sent: Saturday, January 28, 2012 5:43 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones Nice. This should be interesting... On Sat, Jan 28, 2012 at 8:33 PM, David Fotland fotl...@smart-games.com wrote: mfgo1998 is up on kgs if anyone wants to try it. To get a big handicap, set up a 9 stone game and pass until you get up to 29 stones in the same pattern as Martin's game. Regards, David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of David Fotland Sent: Saturday, January 28, 2012 11:34 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones The game with Martin was played on 8/5/1998 at the US Go Congress. My closest source is dated 7/3/1998. I have this built into a gtp front end now, and it's playing different moves from Martin's game, so the engine is not identical. Do you still want to continue with this bet? David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Ingo Althöfer Sent: Thursday, January 26, 2012 1:22 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones Which old Manyface, I have a DOS version from 1990 :) In 1997 ManyFaces 10.0 came to the market (and is sol still nowadays). David Fotland wrote that in the game against Martin Mueller he likely used a source from July 1998. He will try to reanimate that. ingo. -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] minirock2
I tried 3 games with Erica: http://www.gokgs.com/gameArchives.jsp?user=Erica19B In the 1st, 2nd games, Erica was running with 5k playouts/move (1 thread) and Erica lost both games. In the 3rd game, Erica was running with 2 threads and 10s/move (around 15-20k playouts/move). Erica won. My feeling is that Minirock2's playing style is similar to that of Erica. One interesting moment is Black's 13th at C3 in the 2nd game. Minirock'2 C3 is clearly a bad move but it was Erica's ponder hit. Aja -原始郵件- From: Rémi Coulom Sent: Saturday, January 28, 2012 3:23 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] minirock2 I tried 3 games with Crazy Stone: http://www.gokgs.com/gameArchives.jsp?user=bonobot Each time reducing CPU. First game was full power, second game was 4 cores, third game was 2 cores and very fast play. 3rd game was close. Minirock2 looks rather strong. Info now says author is minirock. Maybe we could ask questions to minirock. Rémi On 28 janv. 2012, at 22:30, Hiroshi Yamashita wrote: Sorry, my guess is wrong. I tested 3 games vs Minirock2. Aya won 2 games and lost one. http://www.gokgs.com/gameArchives.jsp?user=ayamc4 But Minirock2 runs on 2 cores 2.8GHz, and Aya is on 6 cores 3.3GHz. I think Minirock2 is 1d or 2d, and if it uses faster machine, 2d or 3d. Zen19 also played 2 games vs Minirock2, and Zen won 2 games. http://www.gokgs.com/gameArchives.jsp?user=zen19 Hiroshi Yamashita - Original Message - From: Erik van der Werf erikvanderw...@gmail.com To: computer-go@dvandva.org Sent: Thursday, January 19, 2012 7:42 AM Subject: Re: [Computer-go] minirock2 No, not Steenvreter. Erik On Mon, Jan 16, 2012 at 12:46 PM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: I thought it is Steenvreter. Hiroshi Yamashita - Original Message - From: Ingo Althöfer 3-hirn-ver...@gmx.de To: computer-go@dvandva.org Sent: Monday, January 16, 2012 6:12 PM Subject: Re: [Computer-go] minirock2 Interesting speculation. Independent of the programmer question I woukd like to see Minirock2 playing rated games. Ingo. Original-Nachricht Datum: Mon, 16 Jan 2012 09:33:32 +0100 Von: Rémi Coulom remi.cou...@free.fr An: computer-go@dvandva.org Betreff: [Computer-go] minirock2 Anybody knows what is minirock2? Was it programmed by rock2? That would be big news :-) Fabien, are you reading? Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Beating old ManyFaces at 29 handicap stones
No, the first game was finished without playing a move. I kept playing pass and at one pass ManyFaces played pass well. Then the game was scored immediately as Black having the whole board, B+327.5. http://files.gokgs.com/games/2012/1/29/ajahuang-mfgo1998.sgf Aja From: Stefan Kaitschick Sent: Saturday, January 28, 2012 9:35 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones Oops, that was the fourth try - the one with the 29.5 point loss. On Sun, Jan 29, 2012 at 4:41 AM, Stefan Kaitschick stefan.kaitsch...@hamburg.de wrote: Awesome attempt! 285 at R8 and you would have crushed it in the first try. Stefan On Sun, Jan 29, 2012 at 2:47 AM, Aja Huang ajahu...@gmail.com wrote: You should let me try. :) Aja From: David Fotland Sent: Saturday, January 28, 2012 6:45 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones I tried 29 stones once and it crushed me J David From: computer-go-boun...@dvandva.org [mailto:computer-go-boun...@dvandva.org] On Behalf Of Michael Williams Sent: Saturday, January 28, 2012 5:43 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones Nice. This should be interesting... On Sat, Jan 28, 2012 at 8:33 PM, David Fotland fotl...@smart-games.com wrote: mfgo1998 is up on kgs if anyone wants to try it. To get a big handicap, set up a 9 stone game and pass until you get up to 29 stones in the same pattern as Martin's game. Regards, David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of David Fotland Sent: Saturday, January 28, 2012 11:34 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones The game with Martin was played on 8/5/1998 at the US Go Congress. My closest source is dated 7/3/1998. I have this built into a gtp front end now, and it's playing different moves from Martin's game, so the engine is not identical. Do you still want to continue with this bet? David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Ingo Althöfer Sent: Thursday, January 26, 2012 1:22 PM To: computer-go@dvandva.org Subject: Re: [Computer-go] Beating old ManyFaces at 29 handicap stones Which old Manyface, I have a DOS version from 1990 :) In 1997 ManyFaces 10.0 came to the market (and is sol still nowadays). David Fotland wrote that in the game against Martin Mueller he likely used a source from July 1998. He will try to reanimate that. ingo. -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] About Minirock
Hi Xaryl, Surprised that Pachi's ladder code can't detect common second line kill like this: .X. XOX ... ### But note in certain tactical situations it’s good to ladder-extend from the second line: http://senseis.xmp.net/?TwoStoneEdgeSqueeze Aja From: Xaryl C Sent: Wednesday, February 15, 2012 4:25 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] About Minirock Yes, it's correct that I use Pachi as foundation and have different playouts and search tree distribution. I don't feel like to submit patches at the moment as the codes are horrendous, but if you want, I can give you compilable codes (but no guarantee it is humanly readable... or can be compiled within 30 seconds :-p) I don't even know if the bot is stronger than Pachi, but I will list some things I noticed about Pachi. Some tests were done some time ago with 9x9 games that I don't remember the detail to well, recent 19x19 tests are done with few playouts per move (4096 per move), and often I move on as long as the change passes binomial test, so I don't have good estimation on elo gain either. In playout policy, 3x3 patterns can use some tactical checks or at least liberty checks; mogo patterns tends to end in crosscuts follow by nonsense too much. There's quite a bit of elo here. The tactical reading codes can improve a bit too, for example, Pachi's ladder code can't detect common second line kill like this: .X. XOX ... ### Heavier tactical readings sometimes don't provide too much elo by itself, but as long as it can probably break even in terms of speed vs elo, I kept them. They might not produce much elo right the way because of the slow down, but it would allow one to add in heavier codes without slowing down the program as much (percentage wise) later on. In playout permits, some moves can be rejected so the program can determine semeai/tsumego/seki better. These moves should be rejected in search tree as well. This might or might not break even in terms of speed vs elo depends on how slow the program currently is, but the effect is very noticeable against human players. Some people kept killing a small group in the beginning in a way that the program thinks it has good chance of living, and then the program would just slack off and lose regardless how many handi it has. Just one warning, be really careful when rejecting moves, or the program will get into situations that it's losing but think it has good chance of winning and never passes or resigns. The dynamic komi scheme works better for me by keeping the winrate between 43% to 48% when losing, and between 50% to 60% when winning. (Values not too carefully tuned, and only on 4096 per move self play even games). 60% feels a bit high but every time I try to decrease it, it seems to lose strength in even game self plays. The max losing komi can also ramp down as the game reaches later stage. Proper dynamic komi provides substantial elo gain at least at self play with small amount of playouts per move. As to the priors, the original program uses somewhat different search tree heuristics, but it doesn't seem to matter much; seems like it's just about biasing the search distribution towards the right direction. However, I don't really like the idea of adding fake winning/losing counts. 5x5 patterns helps a quite a bit even with just hack-ish values. Besides professional games, mid dan games also helps. I am thinking incorporating high kyu games might help as well. I've had my eyes on professor Rémi's 2007 Amsterdam paper since long ago, but never had the time or motivation to try it. Works or not, it's a cool way to determine the values at least :-p Date: Wed, 15 Feb 2012 03:37:07 +0100 From: pa...@ucw.cz To: computer-go@dvandva.org Subject: Re: [Computer-go] About Minirock Hi! On Tue, Feb 14, 2012 at 04:11:01PM +, Xaryl C wrote: I am the one responsible for the bot. I started coding a go bot when school was boring, that was when classical bots prevailed. Then life got busy, I tried MC, made some nice progress in terms of strength, but never even got the chance to properly debug the multi threading codes. Then sometime ago, Minirock (the player) wanted to have a bot, so I hackishly glued (with macros and mass code replacing with scripts) and replaced the playout codes and search distribution along with tactical reading codes and pattern matchers onto Pachi's codes. Great, I'm happy Pachi is useful like this too! If I understand it right, you use Pachi as foundation with custom playout and tactics code? Or is it vice versa? If you have any patches / changes in mind for Pachi (not necessarily with any of your reading code, but even just whatever is needed for better integration of foreign code in Pachi), I will be glad to merge it. Or just if you have any feedback. :-) Best, -- Petr Pasky Baudis Smart data structures and dumb code works a lot better than the other way around. --
Re: [Computer-go] Master Thesis: Multi-Agent Monte Carlo Go
Hi Leandro, Glad to hear your interesting result as well as Zen's great achievement. But the website seems down. Could you please check it? Thanks. Regards, Aja 2012/3/17 Leandro Marcolino soria...@usc.edu Hello all!.. Last year I proposed an improvement over current MCTS algorithms that I called Multi-Agent Monte Carlo Go. My system (a modified version of Fuego) could win about 59% of games against Fuego (in 1000 games) on 9x9 Go, but I don't have results for 19x19 Go... If anyone is interested, you can take a look at the thesis, one paper and the source code at http://teamcore.usc.edu/people/sorianom/mamcgo.html. Regards, Leandro ___ 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] Master Thesis: Multi-Agent Monte Carlo Go
Yes it works now. Thanks. Regards, Aja 2012/3/17 Leandro Marcolino soria...@usc.edu Hello!.. The link seems to be correct... Maybe it is some temporary problem with the server... I could open it right now, could you please try again?.. Thanks!.. Regards, Leandro On Sat, Mar 17, 2012 at 6:54 PM, Aja Huang ajahu...@gmail.com wrote: Hi Leandro, Glad to hear your interesting result as well as Zen's great achievement. But the website seems down. Could you please check it? Thanks. Regards, Aja 2012/3/17 Leandro Marcolino soria...@usc.edu Hello all!.. Last year I proposed an improvement over current MCTS algorithms that I called Multi-Agent Monte Carlo Go. My system (a modified version of Fuego) could win about 59% of games against Fuego (in 1000 games) on 9x9 Go, but I don't have results for 19x19 Go... If anyone is interested, you can take a look at the thesis, one paper and the source code at http://teamcore.usc.edu/people/sorianom/mamcgo.html. Regards, Leandro ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] cgos
On 9x9 Erica now scores about 85% on 300 playouts / move against GnuGo 3.8 Level 10, around of rating 2100. From current CGOS results looks like Crazy Stone is much stronger. :) Aja 2012/4/24 ds d...@physik.de Thanks a lot, there is a long way to go for us with this impressible strength:) Detlef Am Dienstag, den 24.04.2012, 19:43 +0200 schrieb Rémi Coulom: In Crazy Stone, I set the total number of playouts per game, not per move. 250k per game is similar in strength to 2k playouts per move (a bit weaker, maybe). But it takes less time overall for one game (because I apply time-management heuristics to playouts), which allows faster testing. Rémi On 24 avr. 2012, at 18:38, ds wrote: Sorry, I do not want to get all your secrets, but how do you do 250k playouts with seconds for a whole game on all boardsize with 4 cores? Or does reservation mean a minimum reservation? Detlef Am Dienstag, den 24.04.2012, 12:40 +0200 schrieb Rémi Coulom: I reserved a small 4-core node of our cluster to connect 9 fast instances of Crazy Stone, 3 for each board size. On 23 avr. 2012, at 19:48, ds wrote: Hi, would be nice, if one could arrange a typical date, where one can hope to find more programs on cgos. E.g. once a week or once every two weeks on monday? Depending on the opponents one could let the programs run for some hours or some days. For the last weeks it was quite empty with only the standard programs running. You can only get rated around 1800ELO with some accuracy... Detlef ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] A Regression test set for exploring some limitations of current MCTS programs in Go
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=detailaid=3522401group_id=59117atid=489964 https://sourceforge.net/tracker/?func=detailaid=3519829group_id=59117atid=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] A Regression test set for exploring some limitations of current MCTS programs in Go
Hi Jacques, We will appreciate very much if you could participate in our test. In the specification of GTP, about the command 'loadsgf' it says Board size and komi are set to the values given in the sgf file. Board configuration, number of captured stones, and move history are found by replaying the game record up to the position* before move_number *or until the end if omitted. So for the command loadsgf sgf/seki/case1.sgf 4 The program should load the position of case1.sgf BEFORE move 4, not AFTER. Just today some author found a bug of gogui-adapter and kindly reported to me: gogui-adapter incorrectly loads the position AFTER move_number. Markus has already fixed the bug for us, see https://sourceforge.net/tracker/?func=detailaid=3527339group_id=59117atid=489964 If you use gogui-adapter to translate 'loadsgf' for your program, please download the newest version of gogui which is available at https://sourceforge.net/scm/?type=gitgroup_id=59117 Best regards, Aja ___ 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
By the way, to use gogui-adapter to translate 'loadsgf' the command is something like ./run.sh -p java -jar gogui-adapter.jar \PATH_TO_PROGRAM \ -t g_seki_moves.tst (use backslash character (\) to escape the quotes in the string) I used gogui-adapter to run pachi and Mogo as well because they both don't support 'loadsgf'. Please don't hesitate to let me know if it doesn't work for you. Best regards, Aja ___ 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
Hi Olivier, Yes that's our plan. We will appreciate very much if you could participate in our regression test and contribute Mogo's results. It will be interesting to see Mogo's performance of these test cases on large simulations like 1M, 2M, 4M or even 32M over a mega cluster/strong machine. The version of Mogo I ran over the test was downloaded at http://www.lri.fr/~teytaud/mogor It's probably not a current version and I couldn't figure out how to get Mogo's evaluation of a position. Best regards, Aja 2012/5/17 Olivier Teytaud olivier.teyt...@lri.fr If you run tests twice, you get nearly the same results ? Aja: you'll publish results with varying numbers of simulations for MC bots ? Olivier ___ 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
Hi Rémi, Yes, you are right. Case11 is not correct. I have fixed it. Case19 is Hanezeki that might never occur in real games. The purpose of this search is to explore some limitations of current MC Go programs so Martin asked me to design the most difficult seki cases on the earth. Then I just did it. As for komi 7.0, thanks for your suggestion. We will discuss it and announce our decision. Best regards, Aja 2012/5/17 Rémi Coulom remi.cou...@free.fr Now with the correct e-mail address. On 17 mai 2012, at 16:43, Rémi Coulom wrote: I took a closer look at the games. 19 is hanezeki: http://senseis.xmp.net/?Hanezeki I don't worry too much about that. Did this ever occur in a real game? I would recommend using non-integer komi for your tests, because they test the ability of the program to deal with jigo at the same time as they test seki. Dealing with jigo in the search is not an easy job: it is much more difficult to get a consistent search, with proved convergence to optimal play, when the outcome of the game is not binary. Completely greedy search will solve any position with non-integer komi, but it is likely to fail with integer komi (ie, get stuck on jigo when a stronger move can win but has a low evaluation in the beginning of the search). Crazy Stone evaluates hanezeki correctly if komi is set to 7.5 instead of 7.0. Sorry, that should be 6.5. With 6.5, Crazy Stone still fails. So hanezeki is still difficult. Rémi case11 is strange. In the variation contained in the sgf, W loses by two points. Aja, are you sure case11 is correct? Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Another funny kind of seki
Thanks, it is indeed a very interesting seki. In case13 the seki at the bottom-left corner is also formed in a big eye but of a different shape. Aja You'll find in attachment an interesting case of seki that maybe you don't have in your database. The White string in A11 has 3 liberties, but W must not play in any of them, because then A8 and A10 are miai for kill. Black can play in any of them, but search will not play there, because that would make W obviously alive. That position occured in a game that Crazy Stone lost against gnugo. 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] A Regression test set for exploring some limitations of current MCTS programs in Go
Dear all, If you are interested, you can download our latest regression test set at http://webdocs.cs.ualberta.ca/~shihchie/seki-and-two-safe-groups-regression-test.zip which was updated with 1. Newest, bug-free gogui-adapter.jar. 2. Fixed case11.sgf of the seki test set. 3. genmove version of the test files prefixed by g_. We appreciate that not a few authors are interested to participate in our test. Thanks Erik, Yamato and Remi for helping us check and point out the errors in the test set. We will release the final, bug-free version as soon as possible. Best regards, Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] [SPAM] Re: super ko
In case of MCTS, a simple and cheap way is to detect superko as output
filter and use next best move in the tree if superko violation is
detected; however, some situations may be significantly misread because
of that. The other extreme is detecting and avoiding superko even in
simulations; that is fairly expensive (due to a guaranteed cache miss
during hash table lookup, at the very least) and usually *not* done.
The most common compromise is to detect and prune superko moves only
in the game tree and check only the simple ko rule in simulations.
It is not really expensive to detect and avoid superko in simulations. Some
repetitions occur every 4 moves, so you only need to incrementally keep
four keys and compare two keys. For instance,
Key1position before the third last move
Key2position before the second last move
Key3position before the last move
Key4current position
To check super ko for move a in the current position, just do
if (Key1 == ComputeZobristKey(Key4, a))
{
// Forbid move a
}
Aja
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] super ko
Ya, I agree with you all. In fact, four-move cycle is just a basic type of superko, see http://senseis.xmp.net/?Cycle I never managed to prohibit moves that form a cycle of length over six. because I thought cases other than triple ko might occur rarely. But it might be worth a try to handle superko with a hash table, though it might be slow. Aja 2012/6/6 Mark Boon tesujisoftw...@gmail.com On Wed, Jun 6, 2012 at 1:47 PM, Darren Cook dar...@dcook.org wrote: Is four moves the longest super-ko cycle possible? I thought it could be longer, using some capture and refill approach. But I could be very wrong on that. Darren P.S. Well if the bots are psychotic, and because tromp-taylor rules allow suicide, then of course very long cycles are possible, with one player always passing, and the other player filling in the same large eye over and over... My question is more: is a very long cycle possible in a game where both players are playing to win? Four is not the longest, even when you try to win. Quadruple ko takes eight moves to cycle back. Although not common, it can happen. I had it happen in a game in the Dutch championship many years ago. Participation in the World Championship in Japan was on the line, so I can assure you we were both trying to win. Whether it was ideal play to let it happen, that's another question altogether... I don't know what the longest 'reasonable' cycle would be, but I've learned not to put my estimates too low with this type of thing in Go. Before you know it, someone comes up with a 'reasonable' cycle of longer length than you can count. Mark ___ 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] time_left bug of KGS
Hi Nick, I found that KGS sends wrong time_left commands for the byo-yomi time control. For instance, in this TCGA 13x13 computer Go tournament with time setting 0:00+3x0:12(byo-yomi), Fuego received the command from the server time_left w 12 3 which means 12 secs are remained for 3 moves, as the specification of GTP (see http://www.lysator.liu.se/~gunnar/gtp/gtp2-spec-draft2/gtp2-spec.html ) says time_left argument *color time stones* *color* - Color for which the information applies. *time*- Number of seconds remaining. *stones* - Number of stones remaining. But it should be something like time_left w 12 1 because there are 3 periods of 12 secs remained and in each period the engine can fully utilize that 12 secs. As a result of the wrong commands, Fuego abnormally played very fast (2-3 secs per move) and I had to hard-coded Fuego in order to ignore time_left. I know that in the current GTP standard there is no way to describe the byo-yomi time system, but is it a bug of KGS or did I misunderstand something? Thanks, Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Kas Cup
I just realized that it is Lukasz Cup. :) Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Kaś Cup
In such a competition techniques like dynamic komi become extremely important. A program must endeavor to earn more points even if the best result is 0.5 point loss. Interesting. Aja Whether a bot loses to CrazyStone by 10 points or 20 tells us more about its skill than whether it beats break by 100 points or 200. Nick On 10 juil. 2012, at 22:20, Łukasz Lew wrote: Fellow Go enthusiasts, I would like you invite you to: Kaś Cup - a peculiar computer Go tournament. There will be prize pool of total 100$, yay! It will take place on 5th of August on KGS. The peculiarity will come from the scoring method. While this will be a Round Robin, the score for each game won't be +-1 point, but the exact result of the game truncated to the [-50 .. 50] interval. One last rule is that participants may not use more than 4 cores of CPU power. Nick kindly agreed to organize and look after the tournament for which I am grateful. Also he is in charge of choosing a ruleset and time settings. Thank you and let us know if you will participate. Łukasz __**_ 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-gohttp://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-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] Invitation to connect on LinkedIn
LinkedIn I'd like to add you to my professional network on LinkedIn. - Aja Aja Huang Post-Doc at University of Alberta Edmonton, Canada Area Confirm that you know Aja Huang: https://www.linkedin.com/e/8s9wxs-h5de7ee0-5q/isd/8083456779/4mQAfMb2/?hs=falsetok=2HknsJLQ1ZN5k1 -- You are receiving Invitation to Connect emails. Click to unsubscribe: http://www.linkedin.com/e/8s9wxs-h5de7ee0-5q/c5qzyezIDVJogA4XQSHdYeHkQtkUbaMZlBcp/goo/computer-go%40dvandva%2Eorg/20061/I2724790707_1/?hs=falsetok=2inbQu-GlZN5k1 (c) 2012 LinkedIn Corporation. 2029 Stierlin Ct, Mountain View, CA 94043, USA. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] An unusual seki
This seki can be easily handled by forbidding White's self-atari at T9 and S10, which are apparently bad moves because Black's S11 group has a solid real eye at R16. Aja 2012/8/15 Nick Wedd n...@maproom.co.uk 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. I mention this, both to explain to Hiroshi what happened to his bots, and because it is an interesting seki, of I type I had never seen before. Nick (;GM[1]FF[4]CA[UTF-8]AP[**CGoban:3]ST[2] RU[Chinese]SZ[19]HA[5]KM[0.50]**TM[600]OT[5x30 byo-yomi] PW[Blubbel]PB[AyaBot4]WR[3d]**BR[2k]DT[2012-08-14]PC[The KGS Go Server at http://www.gokgs.com/]C[**AyaBot4 http://www.gokgs.com/%5DC%5BAyaBot4[2k\]: GTP Engine for AyaBot4 (black): Aya version 7.42e : If you pass, AyaMC will pass. When AyaMC does not, please remove all dead stones. ] ;B[dd]BL[599.599] ;B[pd]BL[599.278] ;B[jj]BL[598.935] ;B[dp]BL[598.654] ;B[pp]BL[598.306] ;W[qn]WL[599.421] ;B[np]BL[596.887] ;W[qk]WL[598.258] ;B[ip]BL[595.687] ;W[lp]WL[596.32] ;B[kp]BL[594.405] ;W[lo]WL[594.002] ;B[lq]BL[593.188] ;W[mq]WL[592.372] ;B[mp]BL[592.019] ;W[kq]WL[590.811] ;B[lr]BL[591.069] ;W[jp]WL[585.719] ;B[ko]BL[589.498] ;W[jq]WL[576.457] ;B[jo]BL[588.011] ;W[mr]WL[573.518] ;B[ln]BL[586.867] ;W[mo]WL[572.06] ;B[mn]BL[585.458] ;W[no]WL[570.908] ;B[pn]BL[584.089] ;W[oo]WL[567.413] ;B[nn]BL[582.669] ;W[on]WL[560.432] ;B[om]BL[581.529] ;W[po]WL[558.796] ;B[kr]BL[580.282] ;W[jr]WL[555.794] ;B[pm]BL[578.814] ;W[qp]WL[548.661] ;B[qm]BL[577.793] ;W[qh]WL[544.178] ;B[qf]BL[576.209] ;W[io]WL[542.633] ;B[op]BL[574.787] ;W[qo]WL[540.749] ;B[rn]BL[573.321] ;W[ro]WL[538.87] ;B[ok]BL[572.134] ;W[in]WL[536.543] ;B[jm]BL[570.967] ;W[im]WL[534.421] ;B[il]BL[569.256] ;W[jl]WL[531.868] ;B[kl]BL[568.191] ;W[jn]WL[530.255] ;B[kn]BL[566.682] ;W[km]WL[529.03] ;B[ll]BL[565.35] ;W[oh]WL[521.982] ;B[nh]BL[563.782] ;W[og]WL[517.447] ;B[oi]BL[562.486] ;W[ng]WL[515.359] ;B[mg]BL[560.699] ;W[ni]WL[492.114] ;B[mh]BL[559.487] ;W[oj]WL[491.532] ;B[nj]BL[557.992] ;W[pi]WL[488.699] ;B[mi]BL[556.682] ;W[ik]WL[483.385] ;B[hl]BL[555.199] ;W[jk]WL[482.036] ;B[ij]BL[553.922] ;W[hk]WL[480.04] ;B[gl]BL[552.906] ;W[gk]WL[478.407] ;B[fk]BL[551.907] ;W[fl]WL[477.029] ;B[el]BL[550.55] ;W[fm]WL[475.577] ;B[em]BL[549.052] ;W[ek]WL[471.261] ;B[dk]BL[547.847] ;W[fj]WL[469.645] ;B[dj]BL[546.643] ;W[dq]WL[460.631] ;B[cq]BL[545.725] ;W[eq]WL[459.109] ;B[ep]BL[544.085] ;W[cr]WL[457.3] ;B[fp]BL[542.883] ;W[bq]WL[455.966] ;B[cp]BL[541.802] ;W[bp]WL[454.973] ;B[bo]BL[540.752] ;W[gr]WL[452.575] ;B[fn]BL[539.465] ;W[gn]WL[448.989] ;B[hp]BL[537.808] ;W[cf]WL[440.015] ;B[fc]BL[536.774] ;W[bd]WL[436.824] ;B[cc]BL[535.065] ;W[ci]WL[435.404] ;B[di]BL[534.013] ;W[bj]WL[432.975] ;B[gi]BL[532.818] ;W[fi]WL[416.985] ;B[fh]BL[531.688] ;W[gh]WL[416.032] ;B[hi]BL[530.258] ;W[eh]WL[415.031] ;B[fg]BL[529.417] ;W[ei]WL[414.198] ;B[dh]BL[528.432] ;W[eg]WL[412.854] ;B[gg]BL[527.023] ;W[ef]WL[409.783] ;B[cg]BL[525.867] ;W[bg]WL[408.034] ;B[bf]BL[524.529] ;W[bh]WL[403.186] ;B[ce]BL[522.728] ;W[df]WL[401.938] ;B[be]BL[521.808] ;W[cl]WL[398.665] ;B[dl]BL[520.422] ;W[cm]WL[396.907] ;B[dn]BL[519.207] ;W[fo]WL[395.097] ;B[en]BL[517.69] ;W[go]WL[393.291] ;B[br]BL[516.515] ;W[ar]WL[390.882] ;B[dr]BL[515.561] ;W[bs]WL[389.785] ;B[fq]BL[514.28] ;W[er]WL[388.674] ;B[fr]BL[513.124] ;W[ds]WL[387.87] ;B[cn]BL[511.635] ;W[gf]WL[375.574] ;B[bm]BL[510.575] ;W[bl]WL[373.001] ;B[am]BL[509.227] ;W[ai]WL[370.334] ;B[ap]BL[508.188] ;W[hg]WL[366.581] ;B[hh]BL[507.345] ;W[ff]WL[365.669] ;B[gh]BL[506.196] ;W[hf]WL[364.546] ;B[jh]BL[505.012] ;W[nc]WL[360.721] ;B[nd]BL[503.949] ;W[pc]WL[357.991] ;B[qc]BL[502.79] ;W[od]WL[355.262] ;B[oe]BL[501.996] ;W[oc]WL[354.415] ;B[md]BL[500.741] ;W[qd]WL[352.886] ;B[pe]BL[499.497] ;W[qb]WL[349.04] ;B[rc]BL[498.683] ;W[rb]WL[348.013] ;B[rd]BL[498.045] ;W[mc]WL[346.436] ;B[lc]BL[496.798] ;W[lb]WL[342.547] ;B[kc]BL[495.94] ;W[kb]WL[341.515] ;B[jc]BL[494.859] ;W[jb]WL[340.477] ;B[hc]BL[493.773] ;W[ic]WL[339.257] ;B[id]BL[492.944] ;W[ib]WL[338.293] ;B[jf]BL[491.682] ;W[hb]WL[335.898] ;B[gc]BL[490.451] ;W[gp]WL[332.161] ;B[gq]BL[489.594] ;W[ho]WL[328.497] ;B[hq]BL[488.663] ;W[ed]WL[318.979] ;B[hm]BL[487.378] ;W[ec]WL[311.079] ;B[fd]BL[486.128] ;W[db]WL[306.977] ;B[ee]BL[484.995] ;W[de]WL[304.243] ;B[cd]BL[483.928] ;W[fe]WL[301.373] ;B[cb]BL[482.932] ;W[fb]WL[300.273] ;B[pj]BL[482.07] ;W[oi]WL[289.406] ;B[pk]BL[481.069] ;W[qj]WL[287.968] ;B[rl]BL[480.173] ;W[rg]WL[283.271] ;B[ri]BL[479.029] ;W[rk]WL[275.974]
Re: [Computer-go] TAAI 2012 Computer Go Tournaments
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. :) Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Go software on Mac OS X
Dear all, Is there any Go software on OS X like MultiGo on Windows that I can view and edit .sgf game records? Thanks, Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Go software on Mac OS X
2012/11/23 Hideki Kato hideki_ka...@ybb.ne.jp How about qGo (http://qgo.sourceforge.net/)? #I never use it but many Mac users in Japan prefer. Thanks, but the .dmg doesn't work for me. I'm running OS X 10.8.2. It says You can't open the application qGo because PowerPC applications are no longer supported. Cheers, Aja Hideki ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Go software on Mac OS X
2012/11/23 Rémi Coulom remi.cou...@free.fr You can use gogui. Yes, thanks. GoGui seems the best choice. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Go software on Mac OS X
Thanks all. I have successfully viewed Zen's H4 game on Mac OS X. :) Cheers, Aja 2012/11/23 Hideki Kato hideki_ka...@ybb.ne.jp qGo is an open source (GPL) program. Hideki Aja Huang: CALQB9_= 04ex0ksonoj4ecj8hu9kb7q3ts7f34ozymn6nw3d...@mail.gmail.com: 2012/11/23 Hideki Kato hideki_ka...@ybb.ne.jp How about qGo (http://qgo.sourceforge.net/)? #I never use it but many Mac users in Japan prefer. Thanks, but the .dmg doesn't work for me. I'm running OS X 10.8.2. It says You can't open the application qGo because PowerPC applications are no longer supported. Cheers, Aja Hideki inline file ___ 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] Zen resignation positions
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.**htmlhttp://entcog.c.ooco.jp/entcog/event/event20121125.html sgf (player's info updated) http://entcog.c.ooco.jp/**entcog/event/kifu_20121125.ziphttp://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.htmlhttp://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.htmlhttp://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: **313F02D8CF2F41D6AD89D1A42486DD**59@x60: Hi, The game records can be downloaded from http://entcog.c.ooco.jp/** entcog/event/kifu_20121125.lzhhttp://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.ziphttp://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-gohttp://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Hideki Kato mailto:hideki_ka...@ybb.ne.jp** __**_ Computer-go mailing list
Re: [Computer-go] Zen resignation positions
2012/12/2 Erik van der Werf erikvanderw...@gmail.com It doesn't look that easy to me. Have you tried playing it against Erica? No, I haven't set up Erica in this laptop. Another question: how about W H4 instead of H2 for move 28? The corner looks completely alive to me. Even if B has a killing method, it's not easy anyway and that is much better than the unresisting sudden death as Zen played in the game. These 9x9 games, Zen vs. pros, show that LD, like semeai, is still far from an unsolved problem in MCTS. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Multiple unresolved local fights
Hi Ingo, 2012/12/12 Ingo Althöfer 3-hirn-ver...@gmx.de Hi Martin, thanks for the hint. Unfortunately, at the moment I can not download from that site (unexpected error). You could try here https://era.library.ualberta.ca/public/datastream/get/uuid:9e921da9-5176-4327-be53-43b8dec5d1ac/DS1 Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Board Sizes
See http://senseis.xmp.net/?EvenSizedBoards Aja On 7 Jan 2013, at 08:08, Ingo Althöfer 3-hirn-ver...@gmx.de wrote: Go is traditionally played on boards of odd sizes (9x9, 13x13, 19x19, ...) and almost never on even ones (10x10 or 18x18 ...). What are the reasons for this? Ingo (has observed something and wants to put it in context). ___ 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
2013/1/7 David Ongaro david.ong...@hamburg.de Hi, Go really isn't traditionally played on 9x9 and 13x13. They where introduced in recent times to make teaching easier. The traditional board sizes are 17x17 and 19x19. The reason for odd sizes is simple: even sizes lag a tengen which is quite an important point for symmetry and philosophical reasons. A Go board lacking the center of the universe isn't really a universe (= Go Board). Yes, that's an important philosophical reason in ancient China for odd sizes. See (in Chinese) http://www.31v.cn/guojixiangqi/guojixiangqi8183.html Another reason is simply to prevent draws since there was no komi. Agreeing with Don and John, I don't think odd size boards has anything to do with mirror Go. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Crazy Stone!
Congratulations to Crazy Stone! The two games between CS and Zen are quite exciting. Thanks for the report, Nick. I hope you'll get well soon. Aja 2013/1/15 Nick Wedd n...@maproom.co.uk Congratulations to Crazy Stone, undefeated winner of Sunday's 19x19 bot tournament! My (very short) report is at http://www.weddslist.com/kgs/**past/89/index.htmlhttp://www.weddslist.com/kgs/past/89/index.html As usual, I welcome your comments and corrections. 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
Re: [Computer-go] Japanese rules in KGS tournaments
A command something like rule_set japanese might be good. Other rule sets such as Ing and AGA are also popular, see http://senseis.xmp.net/?RulesOfGo Aja 2013/2/23 ds d...@physik.de If my version of the gtp protocol is the latest, there is no command for scoring (it is in the missing feature section). Should we try to extend the specification? Detlef Am Samstag, den 23.02.2013, 13:50 -0500 schrieb Don Dailey: Yes, it would be good to put the Japanese scoring to the test and nothing does this better than a tournament. Don On Sat, Feb 23, 2013 at 1:48 PM, Petr Baudis pa...@ucw.cz wrote: On Sat, Feb 23, 2013 at 07:37:35PM +0100, Erik van der Werf wrote: 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... I second that idea. Having tournaments with Chinese rules by default is friendly to Computer Go beginners, but having a Japanese rules tournament once in a while would be nice to help weed out any bugs (and assess how often some corner cases actually happen :-), especially considering that some important non-KGS tournaments have Japanese rules. -- 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 ___ 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
For the Japanese rules, you can find what I did in Erica at http://www.mail-archive.com/computer-go@dvandva.org/msg00195.html For seki, simply ignore all empty points *connected* with seki groups. Aja 2013/2/22 Martin Mueller mmuel...@ualberta.ca I want to support Japanese rules in Fuego for UEC cup (and for human opponents who prefer these rules). I looked at test cases and previous discussions on the list. My summary so far is on http://sourceforge.net/apps/trac/fuego/wiki/JapaneseRules I did not find any discussions younger than 2009. I also did not find a complete solution. Do people consider this a solved problem? 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] Congratulations to CrazyStone!
Now it seems to me that this is related to the way playouts are done and it will be difficult to improve with Mogo style (rule-based) playouts above certain strength, without using larger patterns and next move choice based on probability distribution. Currently, playing out a simple joseki in a sensible way in simulations will just never happen. This is a bit frustrating since all my attempts at successfully implementing probdist-based playouts have failed so far, but I guess I will just have to try again... To implement softmax, you can refer to my thesis where I have described the framework of the move generator for the playout. Detecting forbidden moves and replacing useless moves by better alternatives are very useful. There you can gain a lot by applying much Go-knowledge. Two good candidate algorithms for training the feature weights are MM and SB(Simulation Balancing). I tried hard but failed to measure any improvement from SB gammas (trained on 9x9) on 19x19. You can use CLOP to tune the MM gammas which are far from optimal according to our experience. Also, my regression test of seki and LD that pachi has participated could be helpful to improve program's tactical strength. In my opinion, that is the most crucial factor to reach high-dan level. Cheers, Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to CrazyStone!
2013/3/8 ds d...@physik.de Can you provide a link to your thesis, as the one I found is dead:) Thanks Detlef http://www.grappa.univ-lille3.fr/~coulom/Aja_PhD_Thesis.pdf Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] UEC Cup day 1
Crazy Stone won. You can see the on-line streaming at http://www.ustream.tv/channel/%E7%AC%AC6%E5%9B%9Euec%E6%9D%AF%E3%82%B3%E3%83%B3%E3%83%94%E3%83%A5%E3%83%BC%E3%82%BF%E5%9B%B2%E7%A2%81%E5%A4%A7%E4%BC%9A Aja 2013/3/17 村松正和 muram...@cs.uec.ac.jp Hi, Here you see the result of the first day of the 6th UEC Cup. http://jsb.cs.uec.ac.jp/~igo/result1.html I'm sorry but the web page is in Japanese. If you click a blue-marked result, you will see the game. (Several games are still in preparation... due to some trouble.) We are also planning to put SGF files of all the games in some place of our web site in a few days. Masakazu Muramatsu 2013/3/17 Aja Huang ajahu...@gmail.com: Thanks Martin. Good luck to 2000-core Fuego. :) Aja 2013/3/16 Martin Mueller mmuel...@ualberta.ca I wrote a brief report for the first day from MP-Fuego's point of view on http://computer-go-adventures.blogspot.com/ I will try to keep posting after each round today. 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] UEC Cup final result and MP-Fuego Nomitan game
Congratulations to Rémi and Crazy Stone! Thanks Martin. The attached game is really funny. Looks like both Fuego and Nomitan overlooked B's T10 and Fuego found it first. Nomitan would have won if it played T17. I'm impressed by Fuego's fighting spirit throughout the whole game. Aja 2013/3/17 Martin Mueller mmuel...@ualberta.ca 1. Crazy Stone 2. Zen 3. Aya 4. Pachi 5. MP-Fuego 6. Nomitan 7. Many Faces of Go 8. blast 9. PerStone 10. MC_ark 11. RGO 12. Izanami 13. GOGATAKI 14. unyugo 15. caren 16. ballade Congratulations to Rémi! MP-Fuego played on 1024 cores of the Hungabee system at University of Alberta today. It beat PerStone in the first round, lost to Zen in the second, won against Many Faces in the 5th-8th place group and beat Nomitan in a wild last game. A few moves before the end, Nomitan had over 90% win rate but it suddenly swung the other way and resigned. At the end there are two semeai involving ko on the board, one of them quite complex, and it is far from clear what would have happened if the game had continued. I attach the game so you can make up your own mind about it :) 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] A Regression test set for exploring some limitations of current MCTS programs in Go
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
Re: [Computer-go] Second Densei-sen results
See the games here with a nice GUI, Ishida Yoshio 9p(W) vs. Zen(B), W+R http://lgs.tw/qyqw7kk Ishida Yoshio 9p(W) vs. CrazyStone(B), B+3.5 http://lgs.tw/q4xo04m Aja 2013/3/20 remi.cou...@free.fr Thanks Petr for your report. Attached is my sgf of the game with Ishida-sensei. Rémi - Mail original - De: Petr Baudis pa...@ucw.cz À: computer-go@dvandva.org Envoyé: Mercredi 20 Mars 2013 13:58:10 Objet: [Computer-go] Second Densei-sen results Hi! On Mon, Mar 18, 2013 at 11:30:05AM +0900, 村松正和 wrote: The results of the second day of the 6th UEC Cup is now on the web at http://www.jsb.cs.uec.ac.jp/~igo/result2.html Today, Zen and CrazyStone both played Ishida Yoshio 9p; there has been maybe about 80-140 spectators in the audience and a professional commentary. Both played with four stone handicap. Zen has lost after fighting hard and ending with a large white group in its moyo that had miai for life that Zen's simulations likely didn't understand; if they did, it seemed like black had some advantage and would have good chances. CrazyStone has played a steady game, resolving an early fight peacefully and entering the yose about 20 points ahead; there was a tricky endgame tesuji at one point that black mishandled, but it didn't matter after all and eventually CrazyStone cruised for a comfortable 3.5pt win. So today's result against Ishida Yoshio 9p on four stones is 1-1! It generally feeled to us that four stones is the correct handicap against professionals at this moment. I don't have CrazyStone's game but I'm attaching Zen's game SGF that I was given by David Fotland. It's a nice testcase. -- 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 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Some CrazyAnalysis of Zen game
In the opening, Zen as usual traded a lot of *cash*(territory) for *a cheque*(the big center). Though this style is far from the current mainstream of the human Go world, it is, however, in accordance with Zen's fighting-oriented playing-style and strategies. Some moves are questionable: N4(72) could be better at P6 or O5, to completely *seal* the breach. O13 instead of G15(74) could consolidate the B's weakness around P6. The 4 stones advantage was completely cancelled when the W group lived in the center since that means Zen's previous efforts were all invalidated. So I think Zen's problem was mainly in the side of game-flow: abandoned territory for the big center, but failed to secure the center. On the contrary, CS secured a lot of territory from the opening. This territory-oriented strategy simplified the game and made the game-flow more stable. Aja 2013/3/20 Ingo Althöfer 3-hirn-ver...@gmx.de Thanks Petr, for the comments. Zen has lost after fighting hard and ending with a large white group in its moyo that had miai for life that Zen's simulations likely didn't understand; if they did, it seemed like black had some advantage and would have good chances. I looked through Zen's game with the analysis tool of CrazyStone. Looking at the histograms (how many random playouts won at which margin) I found some interesting structures: * At move 110 it looked like normally distributed. * At move 120 the development of a right shoulder started. * After 147 for the first time a side maximum occured (around score +79.5) * In the final position, after move 221, crazy shadows were fully developed. It seems that the mailing list does not allowed png attachments. Therefore you have to go to http://www.dgob.de/yabbse/index.php?topic=4956.0 to see diagrams from CrazyAnalysis. Question to stronger go players: Where do you think started Zen's problems? Cheers, Ingo. http://www.althofer.de/crazy-shadows.html ___ 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 Regression test set for exploring some limitations of current MCTS programs in Go
2013/3/26 Hiroshi Yamashita y...@bd.mbn.or.jp My anti-semeai version Aya gets Aya 7/15 46% (anti-semeai) Aya 1/15 6% (normal) But I could not get good result on KGS and selfplay from anti-semeai. Its strength is almost same. Maybe side-effects? One possibility might be your anti-semeai strategy in some situations helps the weaker side in a semeai. For example, suppose B wins by 1 liberty in a semeai. The living percentages of B and W are 80% and 20%. Your anti-semeai strategy might bias the percentages to 70% and 30%. Though both B and W both become *stronger* in semeai by your new rules, but the weaker side (W) gains more. So the *balance* is broken. To handle semeais in playouts, we might need to introduce on-line learning rather than relying on static rules. Aja ___ 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
The attached example shows that a *good rule* might break the balance of playouts and produce a worse evaluation. Suppose in the playout we add a new rule to forbid B's D1 self-atari. This rule makes sense since D1 is a completely meaningless suicide in terms of Go knowledge. But, in fact, this rule might break the balance of the playouts in the sense that B now has a higher probability to live by seki (if W doesn't capture at D1). And B's *live by seki* is a *wrong* result here. The correct result of the playouts should be W kills B with 100% probability. So, along with the rule prohibiting B's self-atari at D1, there should be a new rule to make W capture at D1. Then, the playouts can be balanced toward the correct evaluation. Rules for semeai are usually much more complicated and might break the balance in this way. Aja PlayoutBalance.sgf Description: Binary data ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Using RAVE statistics during playout
To effectively apply RAVE data to the playouts, we must come up with a new idea to incorporate the information of move sequences into RAVE. The main weakness of the AMAF principle is its lack of sequential consideration. And that is NOT an easy problem. In the attached example, W's winning moves are the sequence E9-C9-A9. Unless the playout have some knowledge of semeai, it's hard to help it with solely RAVE statistics: 1. The playouts with W playing at A9 should have 100% winning rate (if W always plays C8 in response to B's B8). Unfortunately, A9 is currently W's illegal move and not considered by RAVE. 2. C9 is W's legal move. The RAVE value of W's C9 could be high since W is able to reach A9 after C9 is played. However, C9 is currently an useless self-atari. It is good only after E9 is played. It is not likely that the standard RAVE could tell us E9 is more urgent than C9, and C9 is more urgent than A9 (by contrast, A9 has the highest winning rate though it's an illegal move), because the statistics are accumulated without considering move sequences. Aja 2013/3/29 Francois van Niekerk flash.sl...@gmail.com The idea sounds pretty much like PoolRave proposed in Biasing Monte-Carlo Simulations through RAVE Values by Rimmel et al. -- Francois van Niekerk Email: flash.sl...@gmail.com | Twitter: @francoisvn Cell: +2784 0350 214 | Website: http://leafcloud.com On 29 March 2013 19:46, Peter Drake dr...@lclark.edu wrote: The Last Good Reply approach is similar (although not identical) to this. We (Orego) got an improvement from it. Some others have, some haven't. https://webdisk.lclark.edu/drake/publications/baier-drake-ieee-2010.pdf On Fri, Mar 29, 2013 at 10:40 AM, Alexander Kozlovsky alexander.kozlov...@gmail.com wrote: Hi! I know that RAVE data typically used during tree traversing. But is it possible to use it during random playout, in order to increase playout quality? On the first sight it seems as dangerous idea, because RAVE statistics are incrementally gathered from the same playouts, and this can lead to problematic positive feedback loop, as in saying The rich get richer and the poor get poorer. That is, random initial fluctuation can get stronger with time and statistics become skewed, because good moves which receive unfortunate initial RAVE data will be ignored in future random playout. But what if we see move selection during random playout as a typical multiarm bandit problem? Then the algorithm of next playout move selection can be the next: 1) select several (say, 4) valid candidate moves for the playout. 2) choose the next move using multiarm bandit formula. We can do this, because for each candidate move we know (a) number of rave wins for this move, (b) number of playouts with this move, (c) total number of playouts (all of this numbers are tied to current UCT node) I think, this should add exploration element to next move selection and prevent skewing of RAVE statistics. I suspect using RAVE data can improve playout strength significantly. Has anybody trying something like this, or it is just crazy idea? ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Peter Drake https://sites.google.com/a/lclark.edu/drake/ ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Weight of moves
1 - finding the local fights causing secondary modes in the distribution by statistical analysis accross terminal nodes 2 - solving the local fight separately 3 - twining the solution (the local game tree) in the global search (game tree) in a way that looks like Conway's methods in Winning Ways. But instead of using he exact methods of Conway, it would be a MC version of it. I think 1 and 3 are the hardest parts to do, but on the other hand, as it is now, this issue is sort of a horizon effect that troubles MC players in general. I think dealing with this will be a major next step in computer go and I'm convinced it can be done, I just don't know how. I agree 1 and 3 are the hardest parts to do, and they are very important. The difficulty is that in the game of Go local fights can correlate with each other at some point. It is not easy even simply to identify the correlation. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Assistance with running Fuego
Hi Brandon, To start Fuego, run the binary *fuego* under /build/opt/fuegomain: .../build/opt/fuegomain$ ./fuego Fuego 1.1.SVN Copyright (C) 2009-2012 by the authors of the Fuego project. This program comes with ABSOLUTELY NO WARRANTY. This is free software and you are welcome to redistribute it under certain conditions. Type `fuego-license' for details. For setting Fuego's parameters, see http://senseis.xmp.net/?Fuego%2FUserManual Cheers, Aja 2013/4/20 Brandon Cieslak bcies...@lclark.edu Hello, Computer Go Mailing List, My programming partner and I are trying to figure out how to run Fuego, and we're having difficulty finding out how to start Fuego from the command line. Could anyone help us out? Thank you very much, Brandon Cieslak ___ 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] Regression Testing
Fuego has lots of regression tests available for many kinds of scenarios. see http://fuego.svn.sourceforge.net/viewvc/fuego/trunk/regression/ Regards, Aja 2013/4/19 Ben Ellis ben.el...@softweyr.co.uk All, Does anyone have a set of SGFs with common use-cases or problem scenarios to test with? At the moment I'm only concerned with rules based tests, i.e. SGFs with several variations of KO (corner, edge, middle), and/or, several variations of Positional/Situational Super KO. I'd also be interested in common AI problems (i.e. simple life death, self filling an eye, KO position, Super KO position, etc) though I'm not that far into my project yet :) Regards, Ben ___ 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] Ranked games on KGS
Probably you have to login as Orego60 and check *Rank* in the profile. Aja 2013/5/12 Peter Drake dr...@lclark.edu I can't figure out what's different. Below is my configuration file at the moment. Any other suggestions? name=Orego60 password=*** room=Computer Go automatch.rank=4k #mode=tournament mode=custom #opponent=mundungus gameNotes=Testing parallel code rules=chinese rules.boardSize=19 #rules.time=180:00 #rules.time=19:00+10/0:30 rules.time=30:00+5x0:30 verbose=t #engine=java -ea -server -Xmx8192M -cp /home/drake/workspace/Orego/bin orego.ui.Orego threads=12 ponder boardsize=13 engine=java -ea -server -Xmx8192M -cp /home/drake/workspace/Orego/bin orego.ui.ClusterOrego localhost=fido.bw01.lclark.eduoutput=/home/drake/tournament_logs/ threads=12 ponder book=FusekiBook grace msec=3000 talk=I am a program and do not talk. reconnect=t undo=f On Sat, May 11, 2013 at 3:16 PM, Hiroshi Yamashita y...@bd.mbn.or.jpwrote: Aya's rate bot config file is like this. Maybe undo setting? And I'm using kgsGtp-3.5.10.zip. --**- engine=c:/go/kgs/AyaBot7/aya.**exe firstargument=--mode gtp verbose=f server.host=goserver.gokgs.com server.port=2379 name=AyaBot7 password=** room=Computer Go mode=custom talk=Good evening. I am a bot and do not talk. gameNotes=AyaMC with 600sim rules=chinese rules.boardSize=19 rules.time=10:00+5x0:30 reconnect=t undo=f --**- Regard, Hiroshi Yamashita - Original Message - From: Peter Drake dr...@lclark.edu To: computer-go@dvandva.org Sent: Sunday, May 12, 2013 6:55 AM Subject: [Computer-go] Ranked games on KGS I've got Orego's new account Orego60 (for using our entire 60-core cluster) authorized as a ranked robot, but it's still playing free games. I logged into the account and checked the rank box. What am I missing? -- Peter Drake https://sites.google.com/a/**lclark.edu/drake/https://sites.google.com/a/lclark.edu/drake/ --**--** __**_ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/**mailman/listinfo/computer-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-gohttp://dvandva.org/cgi-bin/mailman/listinfo/computer-go -- Peter Drake https://sites.google.com/a/lclark.edu/drake/ ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] reusing engine instances in gomill and clop?
map is very slow. Maybe you should use a vector instead. Aja 2013/5/18 ds d...@physik.de Thank you so much! That is what I wanted to do in the first place, but I have problems to do it right. The circular patterns are stored in a std::mapCircPatt , int Now I learned, that c++11 has some class initialization at compile time, but it does not seem to handle such complicated cases?! What I can do is initializing it at startup from vectors (which can be inizialized at compile time), but this is somehow suboptimal, isn't it? But thanks again, for all the helpful comments! Detlef Am Samstag, den 18.05.2013, 09:52 -0700 schrieb David Fotland: Many Faces also compiles the gamma data. It doesn’t read from a file at startup. I prefer to have the engine start each time, since it avoids errors caused by initialization bugs. David -Original Message- From: computer-go-boun...@dvandva.org [mailto:computer-go- boun...@dvandva.org] On Behalf Of Rémi Coulom Sent: Saturday, May 18, 2013 8:47 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] reusing engine instances in gomill and clop? On 18 mai 2013, at 17:30, Matthew Woodcraft wrote: ds wrote: I wonder if anybody had the same problem. We have significant loading time of our go engine (oakfoam due to gamma loading). Both gomill and clop restart the engine for every compatition. In principle it should be possible to reuse the loaded instance? I don't think it would be difficult to teach gomill to reuse an engine process, if you would like that. Do you use the --parallel option? It gets slightly trickier then. Alternatively, does oakfoam have an option to speak GTP by listening on a socket (or something similar like a named pipe)? maybe gogui-server/gogui-client can do that. Also you can compile the startup data into your code to get a fast startup. That's what I do in Crazy Stone. It was necessary for cell phones. But having the program start instantly is also convenient in every day use on a PC. If so, I think it wouldn't be difficult to write a little GTP proxy engine that connects to the socket each time it's run, and passes the commands and responses back and forth. That way you'd solve the problem for CLOP too. -M- ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] reusing engine instances in gomill and clop?
2013/5/18 ds d...@physik.de
Thanks Aja,
in this case I can't I think. Here I look up the large patterns: I could
replace it with some kind of hash table, but this is probably not much
easier to initialize at compile time?
By initializing at compile time, I think they just meant something like
CircPattTable[] = { gamma_1, gamma_2, ..., gamma_n };
In C++ 11, seems you can do
mapCircPatt, int CircPattTable = {{p_1, gamma_1}, {p_2, gamma_2}, ... };
Aja
___
Computer-go mailing list
Computer-go@dvandva.org
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Need help with fuego source code
Hi David, 2013/6/24 David Briemann dbriem...@gmail.com To give you an impression, this is what it looks like in fuego for a well known opening position: http://www.abload.de/img/board7brdj.png So what is puzzling me right now is this: Even if I limit the possible playout moves to the best Y predictions, fuego will play different moves. For example in the linked picture it could play K11, which is not in the predictor move list. This happens too if I disable all heuristics and just do the biased playouts. By play different moves do you mean the moves generated by the tree search or playout policy? From the linked picture it looks like the moves proposed by priors and being searched, since Fuego doesn't consider global patterns playouts in the current version. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Need help with fuego source code
2013/6/24 David Briemann dbriem...@gmail.com I'm beginning to think that I didn't understand the tree search part correctly. You say the tree search generates moves too. I thought moves were only generated in playouts and the tree search part was to follow already played lines until it reaches a position which has not been played out. Probably that's the location were I have too look into. To understand the framework of Fuego, this paper is a good start Fuego - An Open-Source Framework for Board Games and Go Engine Based on Monte Carlo Tree Search http://webdocs.cs.ualberta.ca/~mmueller/ps/fuego-TCIAIG.pdf particularly Section 3 The Fuego Engine, where it describes Fuego's playout policy and the full-board MCTS generator. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] 9x9 result in Olympiad
2013/8/14 Petr Baudis pa...@ucw.cz Aya 1 1 0 0 1 1 1 1 064 Amigo 1 1 1 0 1 1 0 1 064 I'm curious about Amigo - has this program seen any new recent development? As far as I remember, this used to be a rather old and weak program, and its sourceforge page does not show any radical new development, does anyone have any details on this? That was my question as well, and if you ask Wu (the creator of connect-6) the same question he would give exactly the same answer Amigo (or HappyGo) is still a rather old and weak program. :) Recently Amigo successfully scaled up to multiple threads. And that is not only few threads, but hundreds of threads realized by a quite recent technique (or an old one resurrected with a different name?) so called cloud computing which is to distribute different processes (even threads?) to many different nodes, kind of similar to cluster computing. To myself, Amigo is strongly announcing to the whole computer Go community that the era of cloud computing is coming. many details are still not clear, but we can optimistically expect the report of the competition or their coming papers (or the main author Ting-Fu Liao's master thesis http://java.csie.nctu.edu.tw/~tfliao/). Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] On Semeai Detection
I didn't get Don's goodbye message as well, and I found Gmail filtered lots of emails from the list as spams. Aja 2013/10/9 Erik van der Werf erikvanderw...@gmail.com 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 ___ 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 all, 2013/10/10 Lars Schäfers sl...@upb.de 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 If you are interested, the final version of our paper Investigating the Limits of Monte Carlo Tree Search Methods in Computer Go is available for download at https://sites.google.com/site/ajahuangshomepage/home The other paper on the site is MoHex 2.0: a pattern-based MCTS Hex player where we have described the various MCTS enhancements (large patterns in priors, probabilistic simulation, etc) that boosted the MCTS Hex program MoHex for 250-300 Elo, around 80%-85% winning rate against the original version. The latest source of MoHex (hopefully including our C++ implementation of MM) will be publicly available in the near future. Best regards, Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Monte-carlo simulations vs. MCMC
2013/11/1 Rémi Coulom remi.cou...@free.fr In MCMC the distribution is given to you with some kind of mathematical definition, and the challenge is to create a Markov Chain that approximates the distribution well. In MCTS what we really want is a good playout policy and we sample (do playouts) from a given a position with that policy to estimate its value (or expectation). In the context of MCMC seems it is to evaluate a position by doing playouts from another position (and the playouts form a Markov Chain). Aja ___ 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?
2013/11/15 Hiroshi Yamashita y...@bd.mbn.or.jp I agree with you. .OO. OXXO snap back XO.O .XO. For example, Aya often took one O stone before. It is useless. I added the feature Take snap back when there is no ko, its gamma was 0.009, very low. After that, this move is rarely selected in tree. Interesting. In Erica I just completely forbid the capture in playouts. In some cases it's not only about ko but also related to life-and-death. what if the two B stones are the only eye of the surrounding W group? After the capture W will be able to make two eyes and live. In this case, the capture is not a ko threat. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Crazy Stone!
Thanks Nick. I'm impressed by Martin's comments. :) Aja 2013/12/9 Nick Wedd n...@maproom.co.uk Congratulations to Crazy Stone, winner of yesterday's 13x13 KGS bot tournament! My report is at http://www.weddslist.com/kgs/past/98/index.html As usual I will be grateful for your comments and corrections. 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 Crazy Stone!
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 correctly. This game is funny http://files.gokgs.com/games/2013/12/8/AyaMC-DolBaram.sgf I believe such early-ending positions should be handed over to the referee to decide the final result. At least that's what people would do in human Go tournaments. Aja 2013/12/9 Nick Wedd n...@maproom.co.uk Congratulations to Crazy Stone, winner of yesterday's 13x13 KGS bot tournament! My report is at http://www.weddslist.com/kgs/past/98/index.html As usual I will be grateful for your comments and corrections. 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 Crazy Stone!
W can't play N12 if B has more than one external-lib since B will live unconditionally, as the attached example shows. Aja 2013/12/9 Hiroshi Yamashita y...@bd.mbn.or.jp 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 cornerBent4.sgf Description: application/go-sgf ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Crazy Stone!
2013/12/10 Darren Cook dar...@dcook.org If you had four bent-fours, one in each corner, all dead in Japanese rules for the same colour, *and* four un-removeable ko threats, is that the position that will give the biggest difference in score between Japanese and the other scoring methods? Not sure if I understand what you meant here. If there are two groups of corner bent-four on the board, they are both dead because it is the defender that needs ko threats. In the attached example, W needs zero ko threat to kill each B group. W can just eliminate all B's ko threats, then start to kill each group one by one. Aja cornerBent4.sgf Description: application/go-sgf ___ 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 made a test case with four bent-fours and seki. To break the seki at center, at least one of the bent fours must be practically solved. Would be interesting to see a Go program's behavior in this position. :) Aja 2013/12/10 Darren Cook dar...@dcook.org I am afraid can depend on rule set in use. I guess that under Japanese rules bent four is dead by definition. If you had four bent-fours, one in each corner, all dead in Japanese rules for the same colour, *and* four un-removeable ko threats, is that the position that will give the biggest difference in score between Japanese and the other scoring methods? Just curious. (I was going to apologize for being off-topic but, now I think about it, such a position might make a fine corner-case test - in any rule set!) Darren ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go 4BentFours.sgf Description: application/go-sgf ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Anybody going to EGC 2014?
I have registered the weekend tournament ( http://egc2014.com/congress/the-weekend-tournament/). http://egc2014.com/registered-players/ 36 AjaHuang 4d City of London Go Club, http://www.citygoplayers.o I'll offer free, face-to-face teaching games to the interested participants from this list. :) Aja 2014/1/3 Rémi Coulom remi.cou...@free.fr Hi, I am considering the possibility to attend the EGC in Sibiu this summer: http://egc2014.com/ Two weeks is too long for me. I am not familiar with the tournament, but I suppose it is possible to play just a few rounds of the main tournament. And I’ll probably try to participate in a short event, like the 9x9 tournament. There is no computer-go event planned, but if some programmers attend, we can meet and maybe play a few games. Rémi ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to CrazyStone!
Thanks Nick for the report. I'm impressed by DolBaram's performance, in particularly it was using a weaker hardware and scored 1 win, 1 loss and 3 draws against Zen. As you pointed out, seems Zen has a bug on Chinese scoring http://eidogo.com/#url:http://files.gokgs.com/games/2014/2/2/DolBaram-Zen19S-3.sgf Zen (Black) could easily make it a jigo by playing at any point of the the seki area (but W can't) before passing. Aja 2014-02-04 Nick Wedd n...@maproom.co.uk: Congratulations to CrazyStone, winner of the February KGS 9x9 bot tournament! My report is at http://www.weddslist.com/kgs/past/100/index.html As usual I look forward to receiving your corrections and comments. Nick -- Nick Wedd n...@maproom.co.uk ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Japanese rule in MC
Hi Hiroshi,
Maybe it's worth trying never fill own sure territory in the playouts
except replying to opponent's move. That is to say, after certain number
(say 64) of playouts if the territory belongs to one side exclusively (say
99% of the playouts) then from the next playout don't fill the territory in
the playouts actively, unless the opponent plays inside the territory
first.
Aja
2014-03-16 1:25 GMT+00:00 Hiroshi Yamashita y...@bd.mbn.or.jp:
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] Japanese rule in MC
Hmm.. maybe it should be never fill own sure territory in the playouts,
except to defend a threat or reply to opponent's move inside the
territory. This is a hard problem because it involves wasting no move or
cancelling out both side's moves correctly in the playouts.
Aja
2014-03-16 8:28 GMT+00:00 Aja Huang ajahu...@gmail.com:
Hi Hiroshi,
Maybe it's worth trying never fill own sure territory in the playouts
except replying to opponent's move. That is to say, after certain number
(say 64) of playouts if the territory belongs to one side exclusively (say
99% of the playouts) then from the next playout don't fill the territory in
the playouts actively, unless the opponent plays inside the territory
first.
Aja
2014-03-16 1:25 GMT+00:00 Hiroshi Yamashita y...@bd.mbn.or.jp:
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] A few questions from a beginner
Hi Mikko, Welcome to the list. 2014-03-16 18:35 GMT+00:00 Mikko Aarnos mikko.aar...@kolumbus.fi: 1. On single-point eye detection: how is this generally done? I first used the definition that we have a single-point eye if a point has only our stones as neighbours and at most 1 diagonal neighbour is not our stone if in the middle of the board or that all diagonal neighbours are ours if on the edge of the board. This was however very slow, and when I replaced my definition with the definition found in Pachi(all neighbouring stones must be ours, at most 1 enemy stone is our diagonal neighbour if in the middle of the board and none if at the edge of the board) I get 4x the amount of playouts with longer games. Which one is the correct one, or is something else? Pachi's definition looks correct to me. There are a few rare exceptions to this definition but it should work well in general. 2. On playouts: How many light playouts per second on 9x9/19x19 on a single thread is considered fast nowadays? The fastest programs got roughly 80k-100k pure random games per sec on 9x9, and 5k-10k on 19x19, depending on the speed of the CPU. 3. On GTP-protocol and final_score: How do you score a board where the game hasn't ended(i.e. there are moves left which are legal and don't kill off our groups)? Almost all scoring methods require that dead stones are removed, and I can't figure out how to do that easily. Or do you just assume everything on the board is alive? Check out Tromp-Taylor Rules http://senseis.xmp.net/?LogicalRules Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] A few questions from a beginner
The speed of your program would mostly depends on what features you want to support and what data structures you decide to incrementally maintain in the lowest level. Back in 2010, in Erica I got only about 5000-6000 pure random playouts per second on 19x19 at one 2.26 GHz core. For me what really important was the speed of the heavy playouts but of course the light playouts shouldn't be too slow anyway. To learn the tricks (such as using pseudo liberty) for making your program fast, check out Lukasz Lew's libego. Aja 2014-03-20 5:08 GMT+00:00 David Fotland fotl...@smart-games.com: On 9x9 you should be getting about 100k light playouts per second per core on a recent core i7. I was getting almost that many in 2008, and libego was faster. David *From:* computer-go-boun...@dvandva.org [mailto: computer-go-boun...@dvandva.org] *On Behalf Of *Xavier Combelle *Sent:* Wednesday, March 19, 2014 11:44 AM *To:* computer-go@dvandva.org *Subject:* Re: [Computer-go] A few questions from a beginner As other engine: - There is Łukasz Lew. libego https://github.com/lukaszlew/libego (I belived it will give you what you want) - There is oakfoam http://oakfoam.com/ - There is fuego http://fuego.sourceforge.net/ On Wed, Mar 19, 2014 at 5:28 PM, Mikko Aarnos mikko.aar...@kolumbus.fi wrote: Thanks for the info all. Does a fast open-source program exist which uses only light playouts? If there would be one I could get some reliable info on the amount of playouts per second I _should_ be getting. The only open-source I've found is Pachi and that isn't suitable for that purpose. I considered the playout method before asking but I rejected it because I didn't think it was any good at figuring out life/death. Apparently this is not the case so seems like that's what I'm going to be doing. Regards, 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 ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] C++11; threads
Hey Marc, 2014-04-30 8:37 GMT+01:00 Marc Landgraf mahrgel...@gmail.com: Hi, my bot is still under construction, but written entirely under C++11. So few comments: General: Most compilers, especially if you are using Windows, still have problems with C++11 and it's new multithreading library. Right now I'm using mingw-w64-4.8.1 as it has the required support for thread, even so it is done with some workaround via winpthreads, and gives a decently fast code. But I'm also interested if anyone else can share his experience with other compilers. (for windows) Why don't you use Visual Studio 2013? CTP_Nov2013 supports a lot of new C++11 features. http://blogs.msdn.com/b/vcblog/archive/2013/11/18/announcing-the-visual-c-compiler-november-2013-ctp.aspx Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] C++11; threads
I wrote my Go program Erica completely in Visual Studio and had no problem at all. It might be around 20% slower on Windows than on Linux, but compared to other more important factors 20% loss in speed is not really significant. Maybe VS profiler can tell why your program ran awfully slow in debug mode. Aja 2014-04-30 21:38 GMT+01:00 Marc Landgraf mahrgel...@gmail.com: Hey, in the past I tried VS again and again, and in the end always returned back to Code::Blocks... It really feels like VS and me won't find together. Actually, after your comment I tried it again today, but even after spending a decent amount of time of porting it, the program ran awfully slow in debug mode, and crashed, as soon as the VC++ compiler tried to optimize it. (For reasonable performance I need optimization with mingw-w64 as well) Maybe it is just me and my terrible way of coding... But Visual Studio and Visual C++ I can't handle properly. And with Code::Blocks, I fooled around with various versions of GCC, and ended with mingw-w64, which gave me by far the best performance among those supporting the for me relevant C++11-features. Marc 2014-04-30 11:01 GMT+02:00 Aja Huang ajahu...@gmail.com: Hey Marc, 2014-04-30 8:37 GMT+01:00 Marc Landgraf mahrgel...@gmail.com: Hi, my bot is still under construction, but written entirely under C++11. So few comments: General: Most compilers, especially if you are using Windows, still have problems with C++11 and it's new multithreading library. Right now I'm using mingw-w64-4.8.1 as it has the required support for thread, even so it is done with some workaround via winpthreads, and gives a decently fast code. But I'm also interested if anyone else can share his experience with other compilers. (for windows) Why don't you use Visual Studio 2013? CTP_Nov2013 supports a lot of new C++11 features. http://blogs.msdn.com/b/vcblog/archive/2013/11/18/announcing-the-visual-c-compiler-november-2013-ctp.aspx 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] Isn't learning from expert games creating weaknesses?
Hey Marc, It is a common question against supervised learning. I recommend reading about bias-variance dilemma. http://en.wikipedia.org/wiki/Bias-variance_dilemma Aja 2014-05-19 19:10 GMT+01:00 Marc Landgraf mahrgel...@gmail.com: Hi, Today I had an interesting discussion about bots learning from expert (Pro/strong KGS) games to prebias the tree search and/or (soft-)prune parts of the tree. Point was, that playing situational moves out of their usual context can throw the bot off, and force it to 'look' into the wrong direction first. No doubt, the bot can recover from this misjudgement with some playouts, but it is still first send into the wrong direction. Example: Imagine cutting a onespace jump. The bot, looking into it's pattern database, will usually only find this situation, when this move is somehow reasonable. In those cases, often the answer is difficult and sacrifices have to be made. But the most punishing answers won't even be in his database, as he has never seen the case in a pro game, where the move is clearly punishable. But instead the bots tree search will first check the standard answers for difficult cases instead of the clear punishments. It may happen, that the bot then chooses a submissive answer (because that is what usually happens to the reasonable version of the move) instead of the good move/punishment. Surely this example isn't perfect, but I hope it illustrates the problem, I see. Similar things happen with joseki, which can be played correctly, but most likely not properly punished, as the wrong variations are not available in the database, except when they are contextually possible. What makes this problem even worse, is that with the standard methods of playtesting it won't be noticed. In tests against (own or other) engines, if both use a similar database, those moves won't appear out of context. And even playtesting against random opponents on KGS won't show those weaknesses clearly, as even if single players identify those weak spots, their number of games won't be significant usually. I'm not even sure, how one could systematically check for such misjudgements by the bot. Overall, I'm in no way against learning from expert games, and I think there is no doubt, that it is a significant source for improvement of the bots. But the question remains, how those weaknesses could be fixed. The bots have learned how to answer proper play. But how do they learn to answer unusual/bad play? Marc ___ 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!
Thanks Nick. Interestingly, looks like the ranking is strongly correlated with the hardwares. :) 1. Zen19S 50 cores 2. stv 46 threads 3. CrazyStone 24-core 4. Fuego9 12-core 5. AyaMC 6-core NiceGo, GnuGo and MCark were all running on small hardwares. Aja 2014-06-03 15:52 GMT+01:00 Nick Wedd n...@maproom.co.uk: Congratulations to Zen19S, winner of Sunday's 9x9 KGS bot tournament! This was a particularly exciting tournament. My report is at http://www.weddslist.com/kgs/past/103/index.html I hope you will send me your comments and point out my mistakes, as usual. 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] Skip-opening matchmode
2014-06-05 13:54 GMT+01:00 Stefan Kaitschick stefan.kaitsch...@hamburg.de: If a bot is deluded, it will go to work with that delusion on any position. I think it's actually more useful to study positions that are especially susceptible to this behaviour, and then work on remedies such as Remi Couloms simulation balancing. Simulation balancing(SB) was invented by David Silver and Gerald Tesauro, and was published in the world-famous machine learning conference ICML. http://machinelearning.org/archive/icml2009/papers/500.pdf In my PhD career Remi proposed to try it but we certainly didn't invent even modify it. :) Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Specifying Chinese rules in SGF
Hey Peter, To my best knowledge there is no universal standard for that but KGS's convention is probably the most popular for us. http://www.gokgs.com/help/shorthelp.html?helpLocale=de_DE - *Japanese* - mostly used (territory and prisoners count) - *Chinese* - easiest for beginners, bots (programs running on KGS) mostly use chinese rules; area covered by stones plus surrounded territory counts, you may capture closed in stones without changing the result! - *New Zealand* - similar to Chinese, but two or more stones may play self capture. - *AGA* - American Go Accociation, uses japanese and chinese counting (don't mind - the server does the counting for you). Aja 2014-06-23 22:03 GMT+01:00 Peter Drake dr...@lclark.edu: SGF(2) has a place to specify the ruleset, according to http://www.red-bean.com/sgf/properties.html: *Property:* RU*Propvalue:* simpletext http://www.red-bean.com/sgf/sgf4.html#simpletext*Propertytype:* game-info*Function:*Provides the used rules for this game. Because there are many different rules, SGF requires mandatory names only for a small set of well known rule sets. Note: it's beyond the scope of this specification to give an exact specification of these rule sets. Mandatory names for Go (GM[1]): AGA (rules of the American Go Association) GOE (the Ing rules of Goe) Japanese (the Nihon-Kiin rule set) NZ (New Zealand rules) How can we specify the Chinese rules normally used in computer Go? They're *almost* equivalent to AGA rules, but AGA uses situational superko while Chinese uses positional superko. The difference would only come up extremely rarely, potentially making for a nasty bug. For those who don't know these terms: http://senseis.xmp.net/?Superko -- Peter Drake https://sites.google.com/a/lclark.edu/drake/ ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] ieee aticle about computer go by Jonathan Schaeffer
It is a great article overall. I would like it more if it mentions Mogo, at least Follow from the opponent's previous move was actually Mogo's invention in the famous UCT paper, not Fuego's, not to mention a lot of Mogo's achievements on 9x9. But I really like the paragraph describing the great idea RAVE. It might be the first introductory article (for general people) trying to explain RAVE. Aja 2014-07-02 9:11 GMT+01:00 Stefan Kaitschick stefan.kaitsch...@hamburg.de: The artist certainly shows a lack of appreciation and respect for go. Whoever created it, must think that go is already in the bag. Stefan ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] IEEE Spectrum article
Hi Martin, 2014-07-03 22:02 GMT+01:00 Martin Mueller mmuel...@ualberta.ca: We certainly didn’t mean to short-change the MoGo team's or anybody else’s contribution. For this article there were two main points: - try to explain as much as possible how things work in a current program - have some kind of dramatic story for the introduction. We chose the Fuego game against Mr Chou since we are most familiar with it :) Thanks for the explanations. I like this article a lot (even shared it in Facebook), just felt a little bit strange that Mogo is not mentioned at all while the author described the MCTS revolution and techniques. But I completely understand your points: yeah the space is too little to cover other things. It is a great article anyway. Thanks for the write-up. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] [ANN] Imago - Go board optical recognition
I'm dreaming about this scene in EGC 2015.. Several cameras are relaying the games of the best players. A smart optical recognition program automatically converts the streaming images to sgfs and sends them to a Go program. The Go program then shows rich analyses over the games, such as wining rate, best moves, principal variations, estimated territories, etc, even predicts the next moves. A spectator is watching a friend's game and wondering who is ahead. He doesn't understand Go very well. He uses his android phone takes a snapshot. After 3 seconds, the Go software oh his phone immediately tells him that his friend is ahead for 20 points and winning with 90% chance. It's more of just a dream though. :-) Aja 2014-08-12 12:35 GMT+01:00 Petr Baudis pa...@ucw.cz: Hi! Tomas Musil (a student of mine), has created a state-of-the-art open source Go board optical recognition software. We have focused on completely automatic runs, so it automatically detects the board corners and then the stones on the board, and the precision seems pretty good at least in reasonable lighting conditions. You can find it at http://tomasm.cz/imago together with a lot of pictures, documentation and bachelor thesis describing the algorithms in detail. In the thesis, Tomas also compares it against other similar apps, and it appears Imago shows the best performance of all these that were available to us. Unfortunately, we specifically couldn't easily compare it to Remi Coulom's Kifu-snap for multiple reasons - mainly because that is a mobile app. Hopefully, someone will be able to compare these two in the future. At any rate, I think Imago is a great starting point for anyone who would like to play with Go board recognition. My personal dream would be if we added video capability and further improved speed + reliability in time for EGC2015 (in Czech Republic) and were able to deploy it there to transfer large number of top boards. But this will depend on how much time Tomas will have after the summer (and we didn't actually check with EGC2015 organizers yet), so it's still more of just a dream. :-) Petr Baudis ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] [ANN] Imago - Go board optical recognition
2014-08-13 10:02 GMT+01:00 Rémi Coulom remi.cou...@free.fr: There was a bit of irony in Aja’s post, because kifu-snap Crazy Stone already did this in Sibiu. That’s the reason for the last sentence of his message (the one you did not quote) Oh, I actually didn't know your app was already doing that. I thought Crazy Stone was analyzing positions given from sgfs, not snapshots. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] ICGA Journal
I'm a member of ICGA. Yesterday I received the copy of December 2013, about 8 months delay. They can be more efficient in sending the paper journals. :) Aja 2014-08-21 13:31 GMT+01:00 Rémi Coulom remi.cou...@free.fr: I am not a member of the ICGA either. I think it is really ridiculous to have a paper-only computer-science journal in 2014. I’d like to suggest to Detlef to make his paper available online somewhere and post a link to this list. I am looking forward to reading it. Rémi On 21 août 2014, at 14:17, Petr Baudis pa...@ucw.cz wrote: Hi! On Thu, Aug 21, 2014 at 12:44:19PM +0200, Ingo Althöfer wrote: - at least I hope it's still being published, ??? (So you are not a member of the ICGA - otherwise you would know better) Unfortunately not right now, as my Computer Go research profile is very low in recent years. Of course, the ICGA Journal keeps being published. I have all issues until end of 2013. And the March 2014 issue will come soon (delay of 3-5 months is an old ICCA/ICGA Journal tradition). Then I'd suggest that the homepage is updated so that it doesn't list 34-1 from March 2011 as the latest issue (both as direct link and in the contents database). I fear it might be turning some potential authors away. Kind regards, Petr Baudis ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Interviews on codecentric Challenge
Congratulations to Remi and Crazy Stone! Crazy Stone is really strong! Aja -Original Message- From: Rémi Coulom Sent: Sunday, October 05, 2014 11:43 AM To: computer-go@dvandva.org Subject: Re: [Computer-go] Interviews on codecentric Challenge Hi, As promised, here is the analysis of the first game by Crazy Stone: http://remi.coulom.free.fr/CrazyStone/analysis/2014-10-04-fj-CrazyStone/index.html sgf: http://files.gokgs.com/games/2014/10/4/fj-CrazyStone.sgf Thanks FJ, thanks Ingo. I am looking forward to the next game. Rémi On 10/2/2014 5:22 PM, Ingo Althöfer wrote: Hi Nick, thanks for the help. So, I wonder if you can tell me the account names to be used by the players, fj will be Franz-Josef Dickhut CrazyStone will be CrazyStone and what room the games will be played in? The Computer Room. Cheers, Ingo. PS. Please, do not wonder about codecentric with a small c in front. That is official policy of the company. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
[Computer-go] Aya won December 2014 KGS bot tournament
Hi, https://www.gokgs.com/tournEntrants.jsp?sort=sid=933 Congratulations to Aya, the winner of December 2014 KGS bot tournament. I have some questions.(Sorry, Nick, if you would answer any of them in your report but I can't wait anyway) 1. Looks like Aya has improved significantly in the last year. Hiroshi, could you let us know what are the major improvements? 2. Was the new, rewritten version of Orego running on the tournament? 3. Does anyone have information about HiraBot? Is it a new program? It looks very strong, defeating coldmilk and Fuego. I hope this list is not dying. :) Regards, Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
