Re: [computer-go] Two questions to Yamato-san
Yanis Batura wrote: Zen has almost got to 2d level on KGS, http://www.gokgs.com/graphPage.jsp?user=zen19. Unfortunately, it doesn't play since May, 8. 1. Is there any hope that Zen19 will play again on KGS? I'm looking forward to it getting to higher dan levels. When I find a new improvement, Zen will be back to test it. Zen19 might have got 2d then. 2. It's been mentioned earlier that Zen is planned to be released commercially. If yes, when will it be, and what will be the price? Having a dan-level Go advisor running on my home PC at my disposal is a dream of many years! Sorry to keep you waiting. Unfortunately the project is still in the planning stage. I cannot say when or how much it will be. -- Yamato ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Two questions to Yamato-san
2009/5/22 Yamato yamato...@yahoo.co.jp When I find a new improvement, Zen will be back to test it. Zen19 might have got 2d then. Intuition tells me that your Zen might become to Go what Rybka is to chess. Wish you luck in your research, and may the Force be with you! ;) Yanis Batura ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 7x7 komi
On Fri, May 22, 2009 at 2:53 AM, Robert Jasiek jas...@snafu.de wrote: Don Dailey wrote: 7x7 isn't solved by computer, but the best ones play it extrememly well. When looking through sample game trees of small board computer play, my impression was that by far too many trivial moves were not analysed (properly): single passes or seemingly bad plays. When I studied J1989 manually, I had to learn that single passes or seemingly bad plays can be good moves much more often than one fears. Therefore I am extremely sceptical when some claim is made about NxN was solved by strong programs. Usual strength is not a good measure here. Proofs or proof play are appropriate. I haven't heard any claims yet by anyone that 7x7 has been solved. But 5x5 is claimed to have been solved by exhaustive search. And an admissible exhaustive search is a proof. Is the 5x5 claim the one you are skeptical about? Because I am not aware of any non-proof claims. In the case of 7x7 computer players, the practical play is very strong. I think you can construct positions that may be very difficult for computers to solve in 7x7, but from the opening position you are going to find it difficult to find a pathway that fools the computer - if you are on the wrong side of komi. When I did some experiments with Lazarus 2 or 3 years ago, I noticed that the overwhelming majority of games were one-sided when using any 1/2 point komi once you went beyond a reasonable level. As I cranked up the levels, the number of upset loses diminished and I found a level where there were no wins for the wrong side out of perhaps 200 games (don't remember the details.) I don't believe Lazaurs comes close to perfect play on 7x7 but I assume that the 7x7 opening position is relatively easy to win if you are on the correct side of komi.And it may be that the losing side also failed to take advantage of the mistakes of the winning side. So at 8.5 komi black always won, but this does not mean black always played the correct move. - Don -- robert jasiek ___ 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] 7x7 komi
Don Dailey wrote: Is the 5x5 claim the one you are skeptical about? IIRC, I am sceptical about both 5x5 (esp. first move not at tengen) and 6x6. But I do not recall more details. Maybe I have not read all 5x5 papers about claimed solutions. If there is something with mathematical proofs (about the made CG calculations), I would want to read it when I should have time. So far I have seen only Erik's preliminary draft, and that was not on the level of complete mathematical proofs about a solution. -- robert jasiek ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 7x7 komi
2009/5/22 Robert Jasiek jas...@snafu.de: Don Dailey wrote: Is the 5x5 claim the one you are skeptical about? IIRC, I am sceptical about both 5x5 (esp. first move not at tengen) and 6x6. AFAIK the claimed solution is tengen the first move. Maybe you are remebering some interesting lines that starts with (3,2) and (2,2): Subject: computer-go: 5x5 Go is solved Date: Sun, 20 Oct 2002 15:27:04 -0100 From: Erik van der Werf To: COMPUTER GO MAILING LIST Yesterday my program solved 5x5 Go starting with the first move in the centre. As was expected it is a win for the first player with 25 points (the whole board belongs to black). Andrés Domínguez ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 7x7 komi
Robert, A proper search tree is a proof, but there is the issue of determining if there are any flaws in the search and that would require some kind of peer review including a code review. For instance the hash table implementation may not be admissible unless the positions themselves are properly stored which is not the usual technique in game search programs. There is additionally the issue of GHI (Graph History Interaction) which must be handled correctly to consider it a proof.My recollection is that they handled the GHI problem correctly. I'm not sure about hash table signatures. So the game tree searched used in this program (I don't remember who did it) may not be admissible as a proof and I think they actually gave their disclaimers if I remember correctly. Nevertheless, if the program is reporting a win with a very deep search, in my view it's probably correct although it cannot be trusted as an absolute proof. There is the issue of how much we can trust the result of that 5x5 brute force effort. A well engineered brute force alpha/beta search is almost always going to produce correct results - especially if it was designed for that purpose. So personally I'm not heavily skeptical of the conclusions reached although I know there is a some small chance it is wrong. - Don On Fri, May 22, 2009 at 9:49 AM, Robert Jasiek jas...@snafu.de wrote: Don Dailey wrote: Is the 5x5 claim the one you are skeptical about? IIRC, I am sceptical about both 5x5 (esp. first move not at tengen) and 6x6. But I do not recall more details. Maybe I have not read all 5x5 papers about claimed solutions. If there is something with mathematical proofs (about the made CG calculations), I would want to read it when I should have time. So far I have seen only Erik's preliminary draft, and that was not on the level of complete mathematical proofs about a solution. -- robert jasiek ___ 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/
[computer-go] Match: MoGo vs Taranu on 9x9
Today an exhibition match of four games was played between program MoGo and European Champion Catalin Taranu (5p). On 9x9 board, with 30 min for each player. Games (with tons of comments from spectators) can be found in KGS archive, under http://www.gokgs.com/gameArchives.jsp?user=mogoRennes Taranu won the first three games and lost the final one. So, the score was 3-1 for him. Ingo. -- Neu: GMX FreeDSL Komplettanschluss mit DSL 6.000 Flatrate + Telefonanschluss für nur 17,95 Euro/mtl.!* http://portal.gmx.net/de/go/dsl02 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 7x7 komi
Don Dailey wrote: in my view it's probably correct although it cannot be trusted as an absolute proof. A practical computational problem is solved iff a) the underlying theory is published, b) the underlying theory is proven mathematically, c) the algorithm is published, d) the algorithm is proven mathematically, e) the used computer environment is stated, f) there is a statement that the computation has been done successfully and g) it is possible to repeat the computation independently. -- robert jasiek ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 7x7 komi
Robert, Can you dig out the textbook where you got this list from and be more precise about what they are trying to define?It's obvious that they are providing some kind of formal framework for establishing the CREDIBILITY of a claim of proof, not what a proof or solution really is. For instance if I solve a Rubiks cube in private, is it sudenly not solved because I did not follow certain arbitrary formalities?No, it just means that I cannot credibly claim to have solved it. The 5x5 guys, in my opinion, did not credibly solve the 5x5 board. But I don't think they made extravagant claims either. It's extremely common in games research to have papers like this, where results are reported but completely unverifyable and it drives me nuts. - Don On Fri, May 22, 2009 at 2:22 PM, Robert Jasiek jas...@snafu.de wrote: Don Dailey wrote: in my view it's probably correct although it cannot be trusted as an absolute proof. A practical computational problem is solved iff a) the underlying theory is published, b) the underlying theory is proven mathematically, c) the algorithm is published, d) the algorithm is proven mathematically, e) the used computer environment is stated, f) there is a statement that the computation has been done successfully and g) it is possible to repeat the computation independently. -- robert jasiek ___ 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] Match: MoGo vs Taranu on 9x9
http://www.gokgs.com/gameArchives.jsp?user=mogoRennes Taranu won the first three games and lost the final one. So, the score was 3-1 for him. Thanks for the report. As far as I know (not completly sure) it's the first win (in 9x9 game komi 7.5) of a computer against a human as black. I point this out in order to have at least a positive conclusion , as essentially the result is that Catalin won 3 games out of 4 :-) In game 1 mogo lost in the very early moves if I understood well the comments from strong players; in games 2 and 3 mogo lost quite late with some stupid very fast moves - this suggests that perhaps we should save up time in the beginning. Well, it's a conclusion based on a sample of 2 games :-) (by the way we corrected a bug in the time management betweent the 3rd and the 4th game... one day mogo will be bug free :-) ) Best regards, Olivier ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Match: MoGo vs Taranu on 9x9
I think this is still a result to be proud of. In the stupid fast moves, it's correct to play slow in the opening and fast later in the game. It's just a question of how to set the balance. Did you verify that Mogo would have played those stupid fast moves correctly without having to add too much time? - Don 2009/5/22 Olivier Teytaud olivier.teyt...@lri.fr http://www.gokgs.com/gameArchives.jsp?user=mogoRennes Taranu won the first three games and lost the final one. So, the score was 3-1 for him. Thanks for the report. As far as I know (not completly sure) it's the first win (in 9x9 game komi 7.5) of a computer against a human as black. I point this out in order to have at least a positive conclusion , as essentially the result is that Catalin won 3 games out of 4 :-) In game 1 mogo lost in the very early moves if I understood well the comments from strong players; in games 2 and 3 mogo lost quite late with some stupid very fast moves - this suggests that perhaps we should save up time in the beginning. Well, it's a conclusion based on a sample of 2 games :-) (by the way we corrected a bug in the time management betweent the 3rd and the 4th game... one day mogo will be bug free :-) ) Best regards, Olivier ___ 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] Match: MoGo vs Taranu on 9x9
Did you verify that Mogo would have played those stupid fast moves correctly without having to add too much time? No, I've not checked. But the moves were really fast and the comments of humans were in that direction. I agree that we must have more (much more) time for early move. But for the three first games it was really very, very, very slow in the beginning and very, very, very fast in the end - a stupid bug :-/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] Re: verifiable claims
Some lines of play involving large captures will effectively never terminate, even with superko rules in effect. I doubt it is possible to eliminate all these non-terminating lines of play in any way that is provably correct. .. So while claims of solution by exhaustive search might be very convincing, I doubt they can ever be proved. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: verifiable claims
On Fri, May 22, 2009 at 5:47 PM, Dave Dyer dd...@real-me.net wrote: Some lines of play involving large captures will effectively never terminate, even with superko rules in effect. I doubt it is possible to eliminate all these non-terminating lines of play in any way that is provably correct. .. So while claims of solution by exhaustive search might be very convincing, I doubt they can ever be proved. You can just prove that you can make a large-enough chain that is unconditionally alive. I believe that's what Erik did. In practice, you cannot do an exhaustive search using superko rules because then hash table scores cannot be used. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: verifiable claims
On Fri, May 22, 2009 at 5:47 PM, Dave Dyer dd...@real-me.net wrote: Some lines of play involving large captures will effectively never terminate, even with superko rules in effect. But both sides need not play into this to build a proof tree. By the way, an alpha/beta search IS IN FACT a proof, but it has to be constructed properly. I doubt it is possible to eliminate all these non-terminating lines of play in any way that is provably correct. .. So while claims of solution by exhaustive search might be very convincing, I doubt they can ever be proved. Of course it depends on the board size. I think 5x5 can be properly cracked open in the near future. I think the proof tree for both sides can avoid those nearly infinite loops. I do agree that there are some practical difficulties to doing this and being able to claim it's a proof. - 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/
[computer-go] Re: verifiable claims
You can just prove that you can make a large-enough chain that is unconditionally alive. I believe that's what Erik did. In practice, you cannot do an exhaustive search using superko rules because then hash table scores cannot be used. I don't think you can always do that. For example, if white is capturing a chain of size 30, how can you prove that black can never profit by continuing inside that 30 space void. In fact, in many cases you can demonstrate that he should. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: verifiable claims
there are no chains of size 30 on a 5x5 board, and if after a large capture the remaining stones are unconditionally alive the void at the location of the capture cannot be very large. Do remember that we are talking about 5x5 with the first move in the center as the winning move. Cheers, David On 22, May 2009, at 3:15 PM, Dave Dyer wrote: You can just prove that you can make a large-enough chain that is unconditionally alive. I believe that's what Erik did. In practice, you cannot do an exhaustive search using superko rules because then hash table scores cannot be used. I don't think you can always do that. For example, if white is capturing a chain of size 30, how can you prove that black can never profit by continuing inside that 30 space void. In fact, in many cases you can demonstrate that he should. ___ 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/
[computer-go] Re: verifiable claims
At 06:31 PM 5/22/2009, David Doshay wrote: there are no chains of size 30 on a 5x5 board, I'll concede for a 5x5 board, but I think my point is valid for sufficiently large boards, probably 7x7. Almost any strategy other than playing out all legal moves involves a lot of hand waving that is unlikely to be accepted as a proof. There are just too many cases where a pitch inside a captured space has global effects. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: verifiable claims
On Fri, May 22, 2009 at 10:19 PM, Dave Dyer dd...@real-me.net wrote: At 06:31 PM 5/22/2009, David Doshay wrote: there are no chains of size 30 on a 5x5 board, I'll concede for a 5x5 board, but I think my point is valid for sufficiently large boards, probably 7x7. Almost any strategy other than playing out all legal moves involves a lot of hand waving that is unlikely to be accepted as a proof. There are just too many cases where a pitch inside a captured space has global effects. There is no need to explore every cycle to get your proof. I have noticed before that you don't understand how basic alpha beta minimax search works. Here is how it works. Construct an iterative deepening recursive search that returns one of 3 values, it's a win for white, a win for black, or the result is undecided.Do a 1 ply search. If the score returned is -1, the opponent wins and only needed 1 ply. If the score is 1, the side to move wins and only needed 1 ply to do it. If the score is zero, it did not terminate.So you then do a 2 ply search with the same logic. Continue adding an extra ply until the search returns with a non-zero score. We are assuming some half point komi. Ridiculous cycles will be avoided, because the winning player is going to find the shortest proof with iterative deepening. There is no need for the winning side to entertain off-beat lines beyond the current depth goal. You seem to have this idea that because it's possible to have ridiculously long cycles, that they all have to be searched to whatever depths is required to learn the truth. The longest line that needs to be explored is the length of the shortest game that proves who the winner is. Now it could be the case that in some position with a good defense the game can be delayed a few moves with some trick, but it's nothing like the picture you paint where the opponent can take you on infinitely long wild goose chases. You have as much control of the game as your opponent so there is no need to explore a 10,000 move line when a 20 move line wins. You are just thinking about this all wrong and I think you just have a fundamental misconception about how search works. I think with 5x5 this is almost feasible or will shortly be so. Was it Erik who did the 5x5 solver? From what I remember of the paper on this, it sounds like this was pretty close to a proof, or at worst a good outline of how a proving program would work. - 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/
[computer-go] Re: Match: MoGo vs Taranu on 9x9
Olivier Teytaud wrote: http://www.gokgs.com/gameArchives.jsp?user=mogoRennes ... in games 2 and 3 mogo lost quite late with some stupid very fast moves - this suggests that perhaps we should save up time in the beginning. Well, it's a conclusion based on a sample of 2 games :-) I think your sample is already larger than 2 games: look at MoGo's very last 19x19 game in Pamplona (the loss against Zen). I think, in that game too quick play by MoGo in the end was responsible for the loss. http://www.gokgs.com/gameArchives.jsp?user=mogo Mogo[-] Zen19CO [-] 19×19 May 16, 2009, 18:24 (CEST) B+resignation My general impression (also based on experiences from chess): Distributing time rather balanced over the moves is a stable strategy. Of course you will have cases, where 80 seconds instead of 70 seconds make a big difference. But typically this happens much less frequent than differences at for instance 12 sec vs 2 sec. MCTS programs should be more sensible to corruption by small times than (iterative deepening) alpha-beta tree search. Ingo. -- Neu: GMX FreeDSL Komplettanschluss mit DSL 6.000 Flatrate + Telefonanschluss für nur 17,95 Euro/mtl.!* http://portal.gmx.net/de/go/dsl02 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/