Re: [computer-go] Re: 9x9 games wanted
I want to prepare an opening book and I am looking for a 9x9 games collection. So far I have only found in total 244 games, which is for a book much too less (I am used to have the CB-Megabase). Is there a larger collection with at least = 5 Amateur Dan Level available? If the price is reasonable, I am willing to pay for a professionally made collection. I have collected 9x9 games from nngs. http://tusk.dyndns.org/archive/go/1995-2000-go9.zip Regards Martin ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 games wanted and the next big challenge
Except for the relation between not finding 9x9 games which is *not* real go, you can find as many 19x19 games as you want, I agree with Chrilly. Let's accept it. We are amateurs, all except those who are paid by some University to research on go. And even some of them are, because a serious go project takes many years and some have one semester. We have other jobs and (at least myself) try to work less for the money and dedicate 20 hours per week to go programming. We would be very happy to work 60 hours a week on go programming if someone else paid the bills, but that's not the case. I my opinion, the most important software project of the decade, i.e. writing a non-Microsoft _compatible_ operating system, is called wine http://www.winehq.org/ and also looks amateurish. (I don't really know who works there.) 3D studio and other successful projects started as amateur job, so there is nothing wrong in being amateurs. There is no program today which is so much better than free programs that is worth paying for it, so we can't blame the users. We should blame ourselves for not being able to write a program that is worth its price. Also, I don't even doubt that the day computer go can challenge the strongest pro player, the media will understand the importance of the event. (In fact, computer go is already in the media: The Economist, The Times, Scientific American, Abcnews, Reuters, have all written articles in 2007.) And companies will understand that if they want their names related to a historical event like that with no possible repetition in the future, something like the first man on the moon, they will have to pay for it. The money payed for deep blue will be like comparing 1950s with 2007s football contracts. Go is played only by a small freak community. That's not true. Like chess players were admired in the previous century as superintelligent human beings and today no one is interested in chess except the chess community. Go still keeps the supreme form of intelligence myth. And after go, there is void. Of course, you can always invent new games, but you cannot invent millenary games with millions of players. Someone is going to make millions with this. Don't know when, don't know how. I wish I knew ;-) Jacques. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: Congratulations to Crazy Stone!
Hi Nick, thank you for the tournament. I have two questions. One is the start time of the tournament. According to the page: http://www.weddslist.com/kgs/past/index.html, it started 16:00 GMT but it started 24:00 JST (+900). I guess it started 15:00 GMT. #DST problem? The other is about the result. The results on http://www.gokgs.com/tournEntrants.jsp?sort=sid=300 and one on http://www.weddslist.com/kgs/past/28/index.html are different. Which one is correct? # Either is OK for me. -gg Nick Wedd: [EMAIL PROTECTED]: which was the undefeated winner of both divisions in yesterday's KGS bot tournament. My report is at http://www.weddslist.com/kgs/past/28/index.html Nick -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 games wanted and the next big challenge
Very well said Jacques. I agree with everything you said. A couple of comment below. On Mon, 2007-07-09 at 12:02 +0100, Jacques Basaldúa wrote: Except for the relation between not finding 9x9 games which is *not* real go, you can find as many 19x19 games as you want, I agree with Chrilly. I'll bet there have been millions of 9x9 games by very strong players, they are probably just not readily accessible. Let's accept it. We are amateurs, all except those who are paid by some University to research on go. And even some of them are, because a serious go project takes many years and some have one semester. We have other jobs and (at least myself) try to work less for the money and dedicate 20 hours per week to go programming. We would be very happy to work 60 hours a week on go programming if someone else paid the bills, but that's not the case. I my opinion, the most important software project of the decade, i.e. writing a non-Microsoft _compatible_ operating system, is called wine http://www.winehq.org/ and also looks amateurish. (I don't really know who works there.) 3D studio and other successful projects started as amateur job, so there is nothing wrong in being amateurs. I use wine but I personally don't place a great deal of importance on it. I put linux as the single most important amateur project (at least as it started that way) because it is an open source high quality operating system that competes favorably with it's only serious competitor, non-free Unix. Wine is important - no question about it - because the marketing genius behind Windows has created a huge software base. Some of this software is high quality stuff that linux users are even willing to use. There is no program today which is so much better than free programs that is worth paying for it, so we can't blame the users. We should blame ourselves for not being able to write a program that is worth its price. I think computer Go could take off if it were promoted correctly. I don't think it was a complete coincidence that 9x9 computer GO really took off when Nick Wedd starting having monthly computer tournaments and later when CGOS went up.CGOS was created by the computer Go community - a response to a strong desire in the community to have something like it.It provides competition, instant feedback and to a certain extent a sense of status or reward for accomplishing something good. The progress has been enormous in a short time. When CGOS went up I think the strongest program was about 1700 by the CGOS scale. But now 1700 is a pretty mediocre rating on CGOS!I was completely astounded because I did not believe 2000 would be attained any time in the near future - but even 2000 is a modest rating on CGOS now. The progress would still of course be there without CGOS, because the Monte Carlo paradigm was alive before CGOS.But 9x9 would have remained basically unmeasured except in invisible private testing. One might have heard claims of advancements and papers would be written but with such things you almost always have to trust the paper author and his statistics. There is little or no independent verification of results possible. If we want to see rapid 19x19 progress, we need these 3 elements: 1. competition 2. feedback 3. status. This is what something like CGOS provides. The rating and rank provides status and of course the competition is intense and the feedback is instant. Also, it's hard to beg-off when you have something fairly visible like CGOS. If you have a strong commercial program, and you are in the business of making money, it's very tempting to rest on your laurels. You can advertise victories and championships but once you have obtained them, playing in further competitions risks spoiling your reputation (and thus your income.) But with something like CGOS a program like Mogo has bragging rights. It's possible one of the commercial programs is better than Mogo, or perhaps another amateur program is better. But in most peoples minds, Mogo is the best at 9x9 because it was willing to take the risk on CGOS (in all likelihood, it really IS the best and few doubt this.)There are many reasons you might NOT play on CGOS or in tournaments, but most people will probably believe (whether true or not) that you have nothing substantial to show. Of course you simply may not care and that's ok. But you can't make viable claims unless you show up at tournaments, or play on CGOS or in some way take the necessary risks to prove what you have.Tournaments are quite useful and provide visibility and status, but they are infrequent, a very high investment in time and expense for programmers and to be quite frank, they don't really make clear who the best player really is. Any good program has a chance to win a tournament. Here is what we need in order to achieve a Dan level 19x19 player within a couple of years in my
Re: [computer-go] Congratulations to Crazy Stone!
In message [EMAIL PROTECTED], Jason House [EMAIL PROTECTED] writes Open Division Round 1 - You mention AyaBot2 joining it's game with CrazyStone. That should be HBotSVN. Yes, my mistake, now corrected - Printing name and version number happens when the bot crashes, kgsGtp terminates, and a script automatically reconnects the bot, and the cycle repeats. It may be better to say that the SimonBot engine kept crashing when trying to score the game? Yes. Now rewritten as you suggest. Open Division Round 5 - I personally thought the IdiotBot/HBotSVN game had an interesting end position. Despite the extreme weakness of HBotSVN (simply using the UCB algorithm), the position was problematic for several monte carlo engines. HBotSVN thought it had a 100% chance of victory though the bulk of the end of the game because the random playouts assumed idiotbot would fill one of its eyes. I am interested to see that UCB does indeed have that effect. We were discussing it this morning, and wondering if a pure UCB program would be happy about its opponent have a two-point group, expecting one of them to get filled in. But I doubt this would be of general interest to programmers? Also, HB04 does not show up in the names of programs page. Of course, the housebot logins are piling up: HouseBot: Intended for stable version of HouseBot. It's the only ranked account. HB04 - Very old HouseBot 0.4 - Extremely fast play based on 1-ply move heuristics HB05 - HouseBot 0.5 - Global alpha-beta search HBotSVN - Latest and greatest version of HouseBot, generally experimental. Once upon a time, it was version 0.5 . In this last tournament, it was version 0.6. I've added all that now. Thank you for explaining these differences, and for pointing out the other errors. Nick -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Congratulations to Crazy Stone!
In message [EMAIL PROTECTED], Hideki Kato [EMAIL PROTECTED] writes Hi Nick, thank you for the tournament. I have two questions. One is the start time of the tournament. According to the page: http://www.weddslist.com/kgs/past/index.html, it started 16:00 GMT but it started 24:00 JST (+900). I guess it started 15:00 GMT. #DST problem? Yes, my mistake, a DST problem. Some software automatically allows for DS, some automatically converts times to remove the DS correction. It always confuses me. I envy you living in a country where the clocks tell the right time throughout the year. The other is about the result. The results on http://www.gokgs.com/tournEntrants.jsp?sort=sid=300 and one on http://www.weddslist.com/kgs/past/28/index.html are different. Which one is correct? # Either is OK for me. The former is correct, the latter was my error, which I have now corrected. And this was not a copying error by me (as many in the past have been). I assume that I copied the results from the KGS results page after all the games were over, but before I had closed the tournament, so that two of the games had somehow not registered. Thank you for pointing out these errors. Nick -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to Crazy Stone!
On 7/9/07, Nick Wedd [EMAIL PROTECTED] wrote: In message [EMAIL PROTECTED], Jason House [EMAIL PROTECTED] writes Open Division Round 5 - I personally thought the IdiotBot/HBotSVN game had an interesting end position. Despite the extreme weakness of HBotSVN (simply using the UCB algorithm), the position was problematic for several monte carlo engines. HBotSVN thought it had a 100% chance of victory though the bulk of the end of the game because the random playouts assumed idiotbot would fill one of its eyes. I am interested to see that UCB does indeed have that effect. We were discussing it this morning, and wondering if a pure UCB program would be happy about its opponent have a two-point group, expecting one of them to get filled in. But I doubt this would be of general interest to programmers? UCB has nothing to do with it. UCB is simply 1-ply UCT. The interesting part is the anti-eye-filling rule used as part of the MC playouts. As far as I know (and this was mostly confirmed in the computer go chat room), eye-points are detected by the same means. All adjacent points (4 directions) must be the same color, and the surrounding points (8 directions) must have no more than one break, where the edge is included. Here are ALL of the legal eye patterns if you allow them to be rotated. XXX X X XXX XX X X XXX XXO X X XXX |XX | X |XX |XX | X |--- The edge group in the game was something very close to the following: |XXX |X X | XX |XX Note how the lower eye fails to be detected as an eye. There are other examples that can be created that are not detected as eyes, but this is what came up in the game. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to Crazy Stone!
On 7/9/07, Jason House [EMAIL PROTECTED] wrote: On 7/9/07, Nick Wedd [EMAIL PROTECTED] wrote: Also, HB04 does not show up in the names of programs page. Of course, the housebot logins are piling up: HouseBot: Intended for stable version of HouseBot. It's the only ranked account. HB04 - Very old HouseBot 0.4 - Extremely fast play based on 1-ply move heuristics HB05 - HouseBot 0.5 - Global alpha-beta search HBotSVN - Latest and greatest version of HouseBot, generally experimental. Once upon a time, it was version 0.5 . In this last tournament, it was version 0.6. I've added all that now. Thank you for explaining these differences, and for pointing out the other errors. There's still one problem... The description of HB04 says Old version 0.4 of HouseBot. Uses MC/UCB, no UCT. In reality, it does not do MC (or UCB) at all. It votes on moves based on simple heuristics such as descending to protect a group's skirt, placing stones in atari, forming benson unconditional life, etc... It's 1-ply logic like UCB, but has absolutely no MC element. That's why it plays blazingly fast... Games of HB04 vs. IdiotBot have been known to finish in seconds. I should have added a few more things to this... * The UCB version was HBotSVN, but I wouldn't recommend adding that to the description of it. (Actually, the UCB simply represents the configuration for the tournament) * The old entries for the various flavors of HouseBot were left on the bot names page. While I would recommend deleting those old entries, I'd appreciate it if you kept the link to the homepage: http://housebot.sourceforge.net http://housebot.sourceforge.net/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] creating a random position
On 7/9/07, Erik van der Werf [EMAIL PROTECTED] wrote: On 7/9/07, George Dahl [EMAIL PROTECTED] wrote: I think this is what I want. Thanks! So I might have to repeat this a few hundred times to actually get a legal position? Are you aware that nearly all of these positions will be final positions? So I'll repeat my question: why do you need any of this? If you only need final positions it's probably much better to take them from real games, and if you actually need middle game positions you will have to use a different procedure... E. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Won't the final positions be much more likely to be rejected since they are much more likely to be illegal? What is your claim about the distribution of the number of stones on the board with this scheme? I am hoping to use this method to help generate training data for a learning system that learns certain graph properties of the board that can also be computed deterministically from the board position. I know that might sound crazy, but it is working towards the eventual goal of creating feature extractors for Go positions. By learning to map Go positions as an array of stones to Go positions as graphs of strings (instead of just mapping them with a hand coded algorithm) I can take intermediate results in the learner's computation and use it as a feature for another learner. - George ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] 9x9 games wanted and the next big challenge
I'll bet there have been millions of 9x9 games by very strong players, they are probably just not readily accessible. Very unlikely. I'm a strong player (but not very strong - 3 dan amateur), and I've played perhaps a dozen 9x9 games with people who were just learning the rules. I played in a couple of 9x9 tournaments on the crazy go day at the go congress (along with 3-d go, hex go, etc). Most beginners only need a couple of games on 9x9 before they start paying 19x19. 9x9 go is not very interesting to strong players, since it's not really go. I might as well be playing checkers or 9 men's morris :) But with something like CGOS a program like Mogo has bragging rights. It's possible one of the commercial programs is better than Mogo, or perhaps another amateur program is better. But in most peoples minds, Mogo is the best at 9x9 because it was willing to take the risk on CGOS (in all likelihood, it really IS the best and few doubt this.) I can confirm that Mogo is quite a bit stronger than the commercial programs at 9x9 go. I'm not very interested in 9x9 go. Most of the commercial programs have algorithms that don't scale well down to 9x9, since they are all designed for 19x19 go. The 19x19 knowledge that makes them strong does not apply at 9x9. Since people don't play 9x9 go, there is no incentive commercial program authors to make their programs strong at 9x9. Here is what we need in order to achieve a Dan level 19x19 player within a couple of years in my opinion: COMPETITION Not once a year, but constant. A very high profile occasional competition however is still a great and useful thing to have. FEEDBACK You need to always know where you stand so you can constantly be goal oriented. Where you stand in relation to others that is. STATUS There must be some kind of recognition, highly visible acknowledgment of the pecking order to stimulate and motivate the competitors. I agree. Progress was very swift in the Ing competition, with programs improving from about 25 kyu to about 5-8 kyu. Since 2000 the competition and status has been missing, so progress has stopped, or at least is not visible. The algorithms that worked well on a 33 MHz 386 with 0.5 MB memory are very different from what is possible on today's machines. Once money and status come into the picture big time, then cheating will start to play a major role. Cheating did play big role. Even though Ing and FOST had on-site tournaments, there was still the issue of reverse engineering the top programs. I also have to say that Nick Wedd's monthly tournaments are critically important and unquestionably a big part of the sudden progress in computer GO. I think those tournaments and CGOS complement each other in a beautiful way. Probably more credit goes to Nick Wedd's tournaments than CGOS. Those tournament inspired CGOS and they also motivated (in my opinion) a lot of progress in computer chess before CGOS was even up and running.But they do complement each other - CGOS provides instrumentation that KGS is lacking. Nick's tournaments and CGOS have made a huge difference in revitalizing computer go. I'd like to see both expanded to 19x19 with 30 minute per player time limits and some overtime. - Don -David Fotland ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] creating a random position
In that case, you would probably rather have actual Go positions, right? Just grab a bunch of CGOS games (assuming you are studying 9x9) and pick a game and move number at random. On 7/9/07, George Dahl [EMAIL PROTECTED] wrote: On 7/9/07, Erik van der Werf [EMAIL PROTECTED] wrote: On 7/9/07, George Dahl [EMAIL PROTECTED] wrote: I think this is what I want. Thanks! So I might have to repeat this a few hundred times to actually get a legal position? Are you aware that nearly all of these positions will be final positions? So I'll repeat my question: why do you need any of this? If you only need final positions it's probably much better to take them from real games, and if you actually need middle game positions you will have to use a different procedure... E. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Won't the final positions be much more likely to be rejected since they are much more likely to be illegal? What is your claim about the distribution of the number of stones on the board with this scheme? I am hoping to use this method to help generate training data for a learning system that learns certain graph properties of the board that can also be computed deterministically from the board position. I know that might sound crazy, but it is working towards the eventual goal of creating feature extractors for Go positions. By learning to map Go positions as an array of stones to Go positions as graphs of strings (instead of just mapping them with a hand coded algorithm) I can take intermediate results in the learner's computation and use it as a feature for another learner. - George ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] SGF parsing
Do you have a good example of a regular Go game in sgf? A lot of the examples I found on the SGF spec site seem confusing, and not sure if they're even for Go or backgammon, etc. Also is there a command line go conversion program kinda like pgnextract that lets you modify sgf datasets. Like strip comments, etc. All I really want from a program perspective is move lists and user ranking. -Josh On 7/9/07, Don Dailey [EMAIL PROTECTED] wrote: On Mon, 2007-07-09 at 11:49 -0400, Joshua Shriver wrote: I really like the pgn format, just viewing it you can get a feel for what is going on. I tried to figure out the SGF format by looking at it, and have no clue what's going on. SGF has a real grammer associated with it and is technically superior. It provides support for marking up boards and things. However for simple storage requirements, you probably don't need that. However, PGN is wonderfully readable and for what you want more useful in a practical way. Unfortunately, it is not standard for GO and you will be sorry if you want to be able to see a game using a reader for instance. I would stick with SGF for Go. It's isn't that difficult to figure out. You can grab an SGF from somewhere and use it as an example. What we be really useful, is a conversion utility so that you could use both. PGN is more compact but if you zip or compress games it probably doesn't make any difference. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 games wanted and the next big challenge
On 7/9/07, David Fotland [EMAIL PROTECTED] wrote: Very unlikely. I'm a strong player (but not very strong - 3 dan amateur), and I've played perhaps a dozen 9x9 games with people who were just learning the rules. I played in a couple of 9x9 tournaments on the crazy go day at the go congress (along with 3-d go, hex go, etc). Most beginners only need a couple of games on 9x9 before they start paying 19x19. 9x9 go is not very interesting to strong players, since it's not really go. I might as well be playing checkers or 9 men's morris :) I agree that 9x9 is not that interesting for very long, even for beginners, but I'd like to put in a good word here for 13x13. I'm at about 25 kyu on dragongo; nearly all my games are 13x13, and I think I would be having much less fun at 19x19. There seems to be quite a bit of room for strategy at this smaller board size (for example, room enough for joseki patterns, though their significance is probably different) but games are over much quicker, which is an important consideration if you want to have fast games on a non-real-time server. Games take long enough as it is, and quicker feedback is useful for learning. It seems like 13x13 would be a good intermediate step for computer go. - Brian ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] SGF parsing
On Mon, 2007-07-09 at 11:49 -0400, Joshua Shriver wrote: I really like the pgn format, just viewing it you can get a feel for what is going on. I tried to figure out the SGF format by looking at it, and have no clue what's going on. SGF has a real grammer associated with it and is technically superior. It provides support for marking up boards and things. However for simple storage requirements, you probably don't need that. However, PGN is wonderfully readable and for what you want more useful in a practical way. Unfortunately, it is not standard for GO and you will be sorry if you want to be able to see a game using a reader for instance. I would stick with SGF for Go. It's isn't that difficult to figure out. You can grab an SGF from somewhere and use it as an example. What we be really useful, is a conversion utility so that you could use both. PGN is more compact but if you zip or compress games it probably doesn't make any difference. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] 9x9 games wanted and the next big challenge
On Mon, 2007-07-09 at 09:12 -0700, David Fotland wrote: I'll bet there have been millions of 9x9 games by very strong players, they are probably just not readily accessible. Very unlikely. I'm a strong player (but not very strong - 3 dan amateur), and I've played perhaps a dozen 9x9 games with people who were just learning the rules. I played in a couple of 9x9 tournaments on the crazy go day at the go congress (along with 3-d go, hex go, etc). Most beginners only need a couple of games on 9x9 before they start paying 19x19. 9x9 go is not very interesting to strong players, since it's not really go. I might as well be playing checkers or 9 men's morris :) You are in a better position to understand this than I am. I just know I've seen very strong players play 9x9 games on KGS - and over a period of years I would expect there to be a large number although it obviously wouldn't begin to compare to the real game. I don't remember what I actually saw, perhaps it was a match with a weaker player or odds game or something. I also agree that 9x9 doesn't compare to 19x19. I disagree that it's not interesting. It would be uninteresting if, for instance, someone like you were just as good at the top pro's at 9x9. It stops being interested when it can be mastered.If the top players can always play a perfect game, it's not interesting to them, but probably still interesting to me, and to a lesser extent someone like you who would probably be playing close to perfect if the pro's were playing perfect. There would probably be very little difference in someone like you and a top pro and if you played a game well enough you might get some wins if you were on the right side of komi. But this all assumes the game is almost played out. I don't think 9x9 is. I have no argument that any particular individual may not find it interesting as a matter of personal choice. For instance there are many things I don't find interesting even though I haven't mastered them.Or your point of view may be that the bigger board is much MORE interesting, so why bother with smaller ones? But that doesn't take away from the fact that 9x9 is still interesting and still a deep profound game. If you belittle 9x9, indirectly you detract from 19x19 because you imply that the whole game isn't very interesting unless you can put on a massive board. But with something like CGOS a program like Mogo has bragging rights. It's possible one of the commercial programs is better than Mogo, or perhaps another amateur program is better. But in most peoples minds, Mogo is the best at 9x9 because it was willing to take the risk on CGOS (in all likelihood, it really IS the best and few doubt this.) I can confirm that Mogo is quite a bit stronger than the commercial programs at 9x9 go. I'm not very interested in 9x9 go. Most of the commercial programs have algorithms that don't scale well down to 9x9, since they are all designed for 19x19 go. The 19x19 knowledge that makes them strong does not apply at 9x9. Since people don't play 9x9 go, there is no incentive commercial program authors to make their programs strong at 9x9. Most of my comments are directed to 19x19 go. I lay out one possible plan to produce a very strong 19x19 player. I'm interested in 9x9 only as a stepping stone. It's far more manageable and if you can't whip 9x9, you have no chance going bigger. It's way easier to test and get quick results in a methodical way. Here is what we need in order to achieve a Dan level 19x19 player within a couple of years in my opinion: COMPETITION Not once a year, but constant. A very high profile occasional competition however is still a great and useful thing to have. FEEDBACK You need to always know where you stand so you can constantly be goal oriented. Where you stand in relation to others that is. STATUS There must be some kind of recognition, highly visible acknowledgment of the pecking order to stimulate and motivate the competitors. I agree. Progress was very swift in the Ing competition, with programs improving from about 25 kyu to about 5-8 kyu. Since 2000 the competition and status has been missing, so progress has stopped, or at least is not visible. The algorithms that worked well on a 33 MHz 386 with 0.5 MB memory are very different from what is possible on today's machines. Once money and status come into the picture big time, then cheating will start to play a major role. Cheating did play big role. Even though Ing and FOST had on-site tournaments, there was still the issue of reverse engineering the top programs. YES! I remember that and I thought it was a real travesty. I also have to say that Nick Wedd's monthly tournaments are critically important and unquestionably a big part of the sudden progress in computer GO. I think
Re: [computer-go] 9x9 games wanted and the next big challenge
Brian, The idea of moving towards 13x13 appeals to me too. I would even consider removing the 9x9 server and going to 13x13 instead if I didn't think it would cause an out-rage. At some point sticking with 9x9 is going to inhibit progress in my opinion. And a really strong 13x13 program is more likely to be strong at 19x19. - Don On Mon, 2007-07-09 at 09:32 -0700, Brian Slesinsky wrote: On 7/9/07, David Fotland [EMAIL PROTECTED] wrote: Very unlikely. I'm a strong player (but not very strong - 3 dan amateur), and I've played perhaps a dozen 9x9 games with people who were just learning the rules. I played in a couple of 9x9 tournaments on the crazy go day at the go congress (along with 3-d go, hex go, etc). Most beginners only need a couple of games on 9x9 before they start paying 19x19. 9x9 go is not very interesting to strong players, since it's not really go. I might as well be playing checkers or 9 men's morris :) I agree that 9x9 is not that interesting for very long, even for beginners, but I'd like to put in a good word here for 13x13. I'm at about 25 kyu on dragongo; nearly all my games are 13x13, and I think I would be having much less fun at 19x19. There seems to be quite a bit of room for strategy at this smaller board size (for example, room enough for joseki patterns, though their significance is probably different) but games are over much quicker, which is an important consideration if you want to have fast games on a non-real-time server. Games take long enough as it is, and quicker feedback is useful for learning. It seems like 13x13 would be a good intermediate step for computer go. - Brian ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] SGF parsing
Ok found some KGS games, and they make a lot more sense. With the specification I can see what all of the OT, AP, TM, FF, etc commads are. However I don't understand the way it sets the location, so far nothing I've seen describes it. ;B[kr] for example. I thought Go boards used A..x 1..y notation. Perhaps I'm wrong. -Josh On 7/9/07, Joshua Shriver [EMAIL PROTECTED] wrote: Do you have a good example of a regular Go game in sgf? A lot of the examples I found on the SGF spec site seem confusing, and not sure if they're even for Go or backgammon, etc. Also is there a command line go conversion program kinda like pgnextract that lets you modify sgf datasets. Like strip comments, etc. All I really want from a program perspective is move lists and user ranking. -Josh On 7/9/07, Don Dailey [EMAIL PROTECTED] wrote: On Mon, 2007-07-09 at 11:49 -0400, Joshua Shriver wrote: I really like the pgn format, just viewing it you can get a feel for what is going on. I tried to figure out the SGF format by looking at it, and have no clue what's going on. SGF has a real grammer associated with it and is technically superior. It provides support for marking up boards and things. However for simple storage requirements, you probably don't need that. However, PGN is wonderfully readable and for what you want more useful in a practical way. Unfortunately, it is not standard for GO and you will be sorry if you want to be able to see a game using a reader for instance. I would stick with SGF for Go. It's isn't that difficult to figure out. You can grab an SGF from somewhere and use it as an example. What we be really useful, is a conversion utility so that you could use both. PGN is more compact but if you zip or compress games it probably doesn't make any difference. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] creating a random position
On 7/9/07, George Dahl [EMAIL PROTECTED] wrote: On 7/9/07, Erik van der Werf [EMAIL PROTECTED] wrote: On 7/9/07, George Dahl [EMAIL PROTECTED] wrote: I think this is what I want. Thanks! So I might have to repeat this a few hundred times to actually get a legal position? Are you aware that nearly all of these positions will be final positions? So I'll repeat my question: why do you need any of this? If you only need final positions it's probably much better to take them from real games, and if you actually need middle game positions you will have to use a different procedure... E. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Won't the final positions be much more likely to be rejected since they are much more likely to be illegal? Sure, but that does not necessarily matter because there are many more end- than middle-game positions. The reason I brought it up is that I remembered a statement by someone (sorry forgot the source, maybe John or Gunnar remembers) that from all legal positions nearly all can be considered final. Of course one could argue about what makes a position final, obviously not all borders will be nicely closed and generally there will still be some points to be gained, but I think the main idea was that at that point the winner is relatively easy to determine (so one side would normally resign). This also makes sense if you simply look at the expected number of stones on the board. What is your claim about the distribution of the number of stones on the board with this scheme? Simply that for most interesting purposes you will have too many of them on the board. Further, depending on the purpose, one might argue that there are more interesting distributions to sample from (e.g., you could sample from all positions ever played by strong players on KGS). I am hoping to use this method to help generate training data for a learning system that learns certain graph properties of the board that can also be computed deterministically from the board position. I know that might sound crazy, Not to me ;-) but it is working towards the eventual goal of creating feature extractors for Go positions. By learning to map Go positions as an array of stones to Go positions as graphs of strings (instead of just mapping them with a hand coded algorithm) I can take intermediate results in the learner's computation and use it as a feature for another learner. Well, I'm not sure whether this way you will be able to beat hand coded algorithms, but it's certainly interesting to try. In any case I would still think that, to make a strong program, it's better to sample from real games, or maybe do both to see if it makes much difference. Erik ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] SGF parsing
Le lundi 9 juillet 2007 18:46, Joshua Shriver a écrit : Ok found some KGS games, and they make a lot more sense. With the specification I can see what all of the OT, AP, TM, FF, etc commads are. However I don't understand the way it sets the location, so far nothing I've seen describes it. ;B[kr] for example. I thought Go boards used A..x 1..y notation. Perhaps I'm wrong. If i remeber well: * Real life wooden Go board uses A...x without I , 1..y but sgf uses only letters for both coordinates (including I) * Axes are not the same. for board: Ox toward East, Oy toward North for sgf: Ox toward south, Oy toward East Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 games wanted and the next big challenge
I concur with Christian Nilsson; handicap stones permit the win-loss ratio to approximate 50%, where it is more sensitive to improvements. As one tweaks the program, the progress would be measurable within a few games, one's handicap would decrease. Is it possible to tie together the handicap information and the win-loss percentages into a unified ELO-type score? Would an experiment be needed to measure the effect of handicap stones on the probability of winning? Terry McIntyre [EMAIL PROTECTED] They mean to govern well; but they mean to govern. They promise to be kind masters; but they mean to be masters. -- Daniel Webster Park yourself in front of a world of choices in alternative vehicles. Visit the Yahoo! Auto Green Center. http://autos.yahoo.com/green_center/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] SGF parsing
On Mon, 2007-07-09 at 12:46 -0400, Joshua Shriver wrote: Ok found some KGS games, and they make a lot more sense. With the specification I can see what all of the OT, AP, TM, FF, etc commads are. However I don't understand the way it sets the location, so far nothing I've seen describes it. ;B[kr] for example. I thought Go boards used A..x 1..y notation. Perhaps I'm wrong. SGF uses a different coordinate system (making it easier to make mistakes ...) It is all in the Fine Manual: http://www.red-bean.com/sgf/go.html#properties Read it. SGF is surprisingly easy to parse; the only special tokens the parser needs to recognize are ()[]; ( \n and newline escaping add some complexity, but can _initially_ be ignored.) -Josh On 7/9/07, Joshua Shriver [EMAIL PROTECTED] wrote: Do you have a good example of a regular Go game in sgf? A lot of the examples I found on the SGF spec site seem confusing, and not sure if they're even for Go or backgammon, etc. Ignore everything except for GM[1] (= go), and the generic part. For simple sgf, (without variations or game collections) you can create a parser in a few hours. This will probably include reading the manual and understanding the format, too. HTH, AvK ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 games wanted and the next big challenge
I think it would be great to try this out. Perhaps at 13x13. On 7/9/07, Don Dailey [EMAIL PROTECTED] wrote: On Mon, 2007-07-09 at 10:10 -0700, terry mcintyre wrote: I concur with Christian Nilsson; handicap stones permit the win-loss ratio to approximate 50%, where it is more sensitive to improvements. As one tweaks the program, the progress would be measurable within a few games, one's handicap would decrease. Is it possible to tie together the handicap information and the win-loss percentages into a unified ELO-type score? Would an experiment be needed to measure the effect of handicap stones on the probability of winning? I think the common formula is 100 ELO per stone? I think we could start with this guess (or a better one) and after a few weeks of play we could do a statistical analysis to see if things are as they should be. Then we could make any adjustments if needed. CGOS would still use the same scheduling algorithm - trying to prevent serious mismatches. So we would avoid matches that required many stones handicap although they would appear from time to time. The ELO formula is the same. Whatever program is getting the extra stones is assumed (for rating purposes) to be 100H ELO stronger where H is the number of stones handicap. The constant 100 might have to be adjusted of course. It may even be that we have to use a different constant depending where you are at on the ELO scale. With enough games it might be possible to determine if this is needed or not.I've discussed this with Steve in private emails in the past. It might not be difficult to make this auto-adjust. If the server notices that some value isn't predicting the winner very accurately, a tiny adjustment is made. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] the next big challenge - handicap stones on CGOS? or komi?
From: Don Dailey [EMAIL PROTECTED] Is it possible to tie together the handicap information and the win-loss percentages into a unified ELO-type score? Would an experiment be needed to measure the effect of handicap stones on the probability of winning? I think the common formula is 100 ELO per stone? I expect the per-stone adjustment to depend on the size of the board. Two players might be evenly matched on a 19x19 board when the first gets a 9 stone handicap, but first player would have an easy win on a 9x9 board with the same 9 stone handicap. http://senseis.xmp.net/?HandicapForSmallerBoardSizes suggests that 2 stones on 9x9 would be about right; the author also proposes tweaking the komi. According to http://senseis.xmp.net/?AGAHandicaps, the AGA at the 2004 Congress used no handicap stones on 9x9, adjusting the komi instead. Most human players are much more used to handicap stones, but adjusting the komi is another method of compensating for differences in skill, which is much more fine-grained than handicap stones. Got a little couch potato? Check out fun summer activities for kids. http://search.yahoo.com/search?fr=oni_on_mailp=summer+activities+for+kidscs=bz ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Who's going to the Gifu Challenge?
From what I can tell, there has not been a clash of the Go titans since the 2003 Gifu Challenge, which had all of KCC Igo, Haruka, Go+ +, Goemate/Handtalk, Many Faces, GNU Go, and Go Intellect participating. (This was the last public competition for many of these programs.) It seems with the tuning of MoGo and CrazyStone for the full size board and their recent success at the Olympiad, that there is a chance to knock KCC Igo (sold as Silver Star in Japan) from its four year throne. Are any of the Mogo, CrazyStone, and other professional program authors leaving room in their autumn schedules to travel to Ogaki City, Japan for this year's Gifu Challenge? Ian ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 games wanted and the next big challenge
On 7/9/07, Don Dailey [EMAIL PROTECTED] wrote: I think the common formula is 100 ELO per stone? I think we could start with this guess (or a better one) and after a few weeks of play we could do a statistical analysis to see if things are as they should be. Then we could make any adjustments if needed. Once upon a time, people also discussed treating handicapped versions of bots as distinct players with their own ranking. It may be good to experiment with handicap stones that way and then extend it to automatic handicap and folding the results into the rankings. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] creating a random position
If I took a set of game positions, generated by flipping a coin, and generated a histogram of x = black_stones - white_stones I would expect to see the distribution of x looking like a nice Gaussian, centered at zero. If I looked at positions generated by playing out moves, I would expect to see *much* more weight in the tails and the center at a positive value between zero and true Komi. Removing illegal positions from the coin flip set might change the distribution in ways that would be easy to measure and rash to predict. When I look at statistics such as this for real games, I?generally see measurable differences between strong and weak players. - Dave Hillis Check Out the new free AIM(R) Mail -- Unlimited storage and industry-leading spam and email virus protection. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Who's going to the Gifu Challenge?
In message [EMAIL PROTECTED], Ian Osgood [EMAIL PROTECTED] writes From what I can tell, there has not been a clash of the Go titans since the 2003 Gifu Challenge, which had all of KCC Igo, Haruka, Go+ +, Goemate/Handtalk, Many Faces, GNU Go, and Go Intellect participating. (This was the last public competition for many of these programs.) It seems with the tuning of MoGo and CrazyStone for the full size board and their recent success at the Olympiad, that there is a chance to knock KCC Igo (sold as Silver Star in Japan) from its four year throne. Are any of the Mogo, CrazyStone, and other professional program authors leaving room in their autumn schedules to travel to Ogaki City, Japan for this year's Gifu Challenge? Do we know that the Gifu Challenge is going to happen this October? Do you have a URL for it? Nick -- Nick Wedd[EMAIL PROTECTED] ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Who's going to the Gifu Challenge?
- Original Message - From: Ian Osgood [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Monday, July 09, 2007 8:01 PM Subject: [computer-go] Who's going to the Gifu Challenge? From what I can tell, there has not been a clash of the Go titans since the 2003 Gifu Challenge, which had all of KCC Igo, Haruka, Go+ +, Goemate/Handtalk, Many Faces, GNU Go, and Go Intellect participating. (This was the last public competition for many of these programs.) It seems with the tuning of MoGo and CrazyStone for the full size board and their recent success at the Olympiad, that there is a chance to knock KCC Igo (sold as Silver Star in Japan) from its four year throne. Are any of the Mogo, CrazyStone, and other professional program authors leaving room in their autumn schedules to travel to Ogaki City, Japan for this year's Gifu Challenge? Is there a price money? Or at least some sponsorship for traveling? Chrilly ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 games wanted and the next big challenge
Don Dailey wrote: On Mon, 2007-07-09 at 10:10 -0700, terry mcintyre wrote: I concur with Christian Nilsson; handicap stones permit the win-loss ratio to approximate 50%, where it is more sensitive to improvements. As one tweaks the program, the progress would be measurable within a few games, one's handicap would decrease. Is it possible to tie together the handicap information and the win-loss percentages into a unified ELO-type score? Would an experiment be needed to measure the effect of handicap stones on the probability of winning? I think the common formula is 100 ELO per stone? I think we could start with this guess (or a better one) and after a few weeks of play we could do a statistical analysis to see if things are as they should be. Then we could make any adjustments if needed. According to some early experiments I have made on a database of games played by humans on KGS, I'd say it is more likely to be 70 or 80 Elo points. Also, it is likely to depend on strength. I'll be able to give more precise data in a few weeks. The problem with programs is that GNU Go really does not know how to play handicap games. Crazy Stone and, I expect, MC programs in general, should handle handicap much better. Crazy Stone played a few handicap games against weaker humans on KGS two days ago, and it really plays agressive moves when it is behind. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Who's going to the Gifu Challenge?
Ian Osgood wrote: From what I can tell, there has not been a clash of the Go titans since the 2003 Gifu Challenge, which had all of KCC Igo, Haruka, Go++, Goemate/Handtalk, Many Faces, GNU Go, and Go Intellect participating. (This was the last public competition for many of these programs.) It seems with the tuning of MoGo and CrazyStone for the full size board and their recent success at the Olympiad, that there is a chance to knock KCC Igo (sold as Silver Star in Japan) from its four year throne. Are any of the Mogo, CrazyStone, and other professional program authors leaving room in their autumn schedules to travel to Ogaki City, Japan for this year's Gifu Challenge? Ian ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Is there a date set, already ? Anyway, it is not likely I'll go because it usually takes place at the time of year when my classes start. Also I still don't know how to have Crazy Stone play with Japanese rules. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Who's going to the Gifu Challenge?
On Jul 9, 2007, at 11:17 AM, Nick Wedd wrote: In message [EMAIL PROTECTED], Ian Osgood [EMAIL PROTECTED] writes From what I can tell, there has not been a clash of the Go titans since the 2003 Gifu Challenge, which had all of KCC Igo, Haruka, Go + +, Goemate/Handtalk, Many Faces, GNU Go, and Go Intellect participating. (This was the last public competition for many of these programs.) It seems with the tuning of MoGo and CrazyStone for the full size board and their recent success at the Olympiad, that there is a chance to knock KCC Igo (sold as Silver Star in Japan) from its four year throne. Are any of the Mogo, CrazyStone, and other professional program authors leaving room in their autumn schedules to travel to Ogaki City, Japan for this year's Gifu Challenge? Do we know that the Gifu Challenge is going to happen this October? Do you have a URL for it? Nick -- Nick Wedd[EMAIL PROTECTED] I guess there has not been an announcement yet, though it has been in early October the previous two years. Last year's pre-announcement was sent to the list on July 5. Perhaps the manager of the Computer Go Forum (http://www.computer-go.jp/index.html) has more information. Ian ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Who's going to the Gifu Challenge?
There is prize money. I think it was about $3000 US last year for first place. No remote computing, so if like me you use a cluster, you must bring it. Cheers, David On 9, Jul 2007, at 11:33 AM, chrilly wrote: travel to Ogaki City, Japan for this year's Gifu Challenge? Is there a price money? Or at least some sponsorship for traveling? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] SGF parsing
Yes, without variations SGF is not hard. Unfortunately, doing it right when you want to look at lots of variations at each move is quite tricky. We need to do this to inspect what SlugGo is considering on each of the many CPUs we are using, and every now and again we need to revisit this code. Cheers, David On 9, Jul 2007, at 10:31 AM, Unknown wrote: For simple sgf, (without variations or game collections) you can create a parser in a few hours. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Cgos problems and resignation
Don wrote: On Sun, 2007-07-08 at 12:53 +0200, Magnus Persson wrote: I just had an exception in Valkyria because it recieved play b resign from the server. As far as I know CGOS used to to send nothing to the winner when a program resigned. Am I wrong or has this something to do with the current stability problems of CGOS? Her is the end of the log. Game was 69653 12:35:43C-E genmove w 12:35:45E-C = j1 12:35:45C-S j1 12:35:47S-C play b RESIGN 47144 12:35:47C-E play b RESIGN It's easy to take this out of the client. I think I have it in because RESIGN is a legitimate GTP move and part of the standard. [...] It's not. Resign is a valid response to the genmove command but it's not a legitimate move and cannot be used with the play command. /Gunnar ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Cgos problems and resignation
Ok, my bad.I will take it out of the next client version. If it causes anyone trouble it can easily be removed from the client, just let me know. - Don On Mon, 2007-07-09 at 23:24 +0200, Gunnar Farnebäck wrote: Don wrote: On Sun, 2007-07-08 at 12:53 +0200, Magnus Persson wrote: I just had an exception in Valkyria because it recieved play b resign from the server. As far as I know CGOS used to to send nothing to the winner when a program resigned. Am I wrong or has this something to do with the current stability problems of CGOS? Her is the end of the log. Game was 69653 12:35:43C-E genmove w 12:35:45E-C = j1 12:35:45C-S j1 12:35:47S-C play b RESIGN 47144 12:35:47C-E play b RESIGN It's easy to take this out of the client. I think I have it in because RESIGN is a legitimate GTP move and part of the standard. [...] It's not. Resign is a valid response to the genmove command but it's not a legitimate move and cannot be used with the play command. /Gunnar ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Explanation to MoGo paper wanted.
Dave wrote: We have seen a similar effect many times in MoGo. Often we try something that seems like it should improve the quality of the simulation player, but it makes the overall performance worse. It is frustrating and surprising! Has anyone else encountered this? I'm not surprised. The goal of Monte Carlo simulations should be to provide an unbiased estimate of the true min-max value with as low variance as possible. This has little to do with strength, unless you happen to find a perfect simulation player, but then the whole search business becomes moot. The fact that many modifications of uniformly random playouts simultaneously improve simulation playing strength and overall strength is a red herring. Uniformly random playouts are strongly biased to overestimate the value of having tightly connected stones since e.g. one space jumps become cut through disproportionally often compared to what happens in relevant paths through the min-max tree. Almost any change in simulation policy that counters this tendency will improve overall strength and likewise pretty much every sensible change will improve simulation strength compared to uniformly random play. At higher levels something that may happen is that a change in the simulation policy improves the skill at making life in tight spots, without changing other skills. This would likely improve simulation strength but would cause a bias for positions where there's room for a futile invasion that barely fails, decreasing overall strength. Similar phenomena have turned up in GNU Go over the years. If you tune tactical reading or life and death reading to find some new class of attacking moves, results are likely to become worse if you don't do matching changes in the capability to find defense moves. There's also the classical effect of fixing an obvious mistake just to find some regression tests starting to fail. Closer examination shows that the tests were previously only passing because there were two mistakes that cancelled each other and fixing one of them breaks the balance. /Gunnar ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Explanation to MoGo paper wanted.
Don wrote: Of course now we just had to go and spoil it all by imposing domain specific rules. I have done the same and I admit it.It would be fun to see how far we could go if domain specific knowledge was forbidden as an experiment. Once patterns are introduced along with other direct Go knowledge, it's still fun but it feels a bit wrong, kind of like cheating. It's clear that when we do this, we introduce strengths and weaknesses to the program, making it a bit more fragile, less universal or robust. Stronger too, but more susceptible to in-transitivity. I'm on the other side of this issue. In my opinion all kinds of go knowledge are fair game and I'm rather disappointed that so small amounts of domain specific knowledge have been merged with the UCT search approaches. /Gunnar ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 9x9 games wanted and the next big challenge
Yes. This number is strongly dependent on strength and board size I think. Very roughly speaking, you can argue as follows 1) in a 9x9 game, the weaker player has only 1/4 as many moves in which to throw away the handicap advantage (compared to 19x19). 2) weak players lose so many points compared to perfect play that the final score (the difference between the number of points the two players lose) has a large variance compared to the value of a handicap stone. According to some early experiments I have made on a database of games played by humans on KGS, I'd say it is more likely to be 70 or 80 Elo points. Also, it is likely to depend on strength. I'll be able to give more precise data in a few weeks. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Explanation to MoGo paper wanted.
Benjamin wrote: I have build just for fun a simple BackGammon engine. [...] Interesting - did you also try it for chess, or do you think there's no point in this? This is a bit of speculation since I don't know enough about chess but I suspect that uniform random simulation in go is about as reliable an evaluation as a plain counting of piece values in chess, except that the latter comes without noise. So doing random simulations in chess would only make life more difficult, unless possibly if the simulation policy was really good. Doing UCT search instead of alpha-beta with some deterministic evaluation function might be an interesting experiment but I suspect alpha-beta is more efficient in that case. Othello seems like a better fit for UCT/MC and I suppose that has already been tested. /Gunnar ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] creating a random position
On 7/9/07, Gunnar Farnebäck [EMAIL PROTECTED] wrote: Erik wrote: Sure, but that does not necessarily matter because there are many more end- than middle-game positions. The reason I brought it up is that I remembered a statement by someone (sorry forgot the source, maybe John or Gunnar remembers) that from all legal positions nearly all can be considered final. Of course one could argue about what makes a position final, obviously not all borders will be nicely closed and generally there will still be some points to be gained, but I think the main idea was that at that point the winner is relatively easy to determine (so one side would normally resign). This also makes sense if you simply look at the expected number of stones on the board. I don't remember that statement. My guess, without trying it out, is that most legal positions are rather unsettled and that the number of positions that are final in any strong sense is a tiny fraction of the legal positions. Ok, then probably I'm mistaken and read it in a different context. In any case, my statement should be relatively easy to falsify; just generate some positions and count the fraction that is easily solved. If correct, a decent uct search, or maybe even a traditional solver, would in most cases quickly converge to an extreme probability of winning. Erik ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Explanation to MoGo paper wanted.
This discussion reminds me of a naive theory that I sometimes wonder about: Since the players in the playouts are so weak, it seems like the improving the ability to defend a strong position from a not-very-clever move (and not lose it via a blunder) should be more important than improving the ability to find an attack. If there are two equally bad players that can easily attack each other but can't defend, it seems like the results will be close to random, almost regardless of starting position, unless it is very strong. On the other hand, if two bad players are somewhat better at defense but lousy at seeing weaknesses in the other side, there will be less noise and the one with more territory will tend to win, but an attack on a mostly solid position will sometimes be found via a random move, and given enough playouts, this will result in the probability of defense with a weakness being slightly lower than a truly winning position. It seems like this effect would be especially true of the endgame where there aren't so many points to take, but a position could be lost due to a blunder. I'm not sure how useful that is, since to defend a position you need to know how it might be attacked, but perhaps it leads somewhere? - Brian ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/