Re: [computer-go] Interesting problem
On 31-dec-06, at 15:34, David Fotland wrote: A strong chinese player using chinese rules will pick up a point or two during the dame filling stage when playing a strong japanese player. The Chiense player will choose earlier moves that gain a later dame point that the japanese player will think have no benefit over other moves. I'm rather late to the discussion, having been on vacation, but the above seems strange. Choosing a move that will gain a later dame point is equivalent to making a point. Therefore by definition the move the chinese player made was not a dame point. As long as both players fill in dame, the Chinese player will never gain a point. The only case where I've seen strong Chinese players gain a point very late in the game against other strong players is by winning the last half-point ko but instead of filling it he fills another dame-point. Provided he has many more ko-threats so he can wait filling the ko until all dame are filled he gains either zero or two points (depending on whether the number of remaining dame was even or uneven.). Mark ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] Interesting problem
I disagree that in Chinese rules a player can afford to play unnecessary defensive moves inside his territory. If you play an unneccessary move inside your territory while there are still dame points, you will lose points under Chinese rules, because your move has no value, but your opponent will fill a dame which has value. So the only way you can play inside of your territory in Chinese rules whithout losing points is to do it after all dame points are filled. You would have to postpone your potentially neccessary defensive moves until all damepoints are filled, which is usually too late if there really was a threat. My conclusion is that there is hardly any difference between Chinese and Japanese rules in penalizing unneccessary defensive moves inside you territory. There is only one useful moment under Chinese rules where you can play an unneccessary defensive move whithout losing points: just after your opponent filled the last dame. Playing defensive moves after your opponent passes would cost nothing in Chinese rules, but i don't see any usefullness in this. Dave - Oorspronkelijk bericht - Van: David Fotland <[EMAIL PROTECTED]> Datum: zondag, december 31, 2006 6:34 pm Onderwerp: RE: [computer-go] Interesting problem > People who play by Japanese rules fill the dame before passing and > scoring.Professional game records leave those moves out since they > are irrelevant, > but if you go to a club and watch people playing, they usually > fill the dame > before passing. Sometimes you will see a verbal agreement that > the game is > over and both players will fill dame at the same time. > > A strong chinese player using chinese rules will pick up a point > or two > during the dame filling stage when playing a strong japanese > player. The > Chiense player will choose earlier moves that gain a later dame > point that > the japanese player will think have no benefit over other moves. > > Japanese rules has the minor advantage that they penalize plays > into your > opponent's terrioty that are not answered, so it is more risky in > Japaneserules to make an unsound invasion at the end of the game. > Chinese rules > tolerates making these invasions and making safety moves to ensure > aen enemy > group is dead, or even taken off the board. This is why computer > competitions like chinese rules. Japanese rules make the game more > difficult. > > Chinese rules have the advantage of being more elegant. Disputes > over group > status can be played out. Japanese rules require agreement on > group status, > and there are rare situations that are very difficult to resolve. > > Most of the world plays be Japanese rules, so any commercial > program must > implement Japanese rules. > > David > > > -Original Message- > > From: [EMAIL PROTECTED] > > [mailto:[EMAIL PROTECTED] On Behalf Of Don Dailey > > Sent: Sunday, December 31, 2006 7:08 AM > > To: computer-go > > Subject: Re: [computer-go] Interesting problem > > > > > > On Sun, 2006-12-31 at 13:00 +, Jacques Basaldúa wrote: > > > I don't agree on that. If you are used to Chinese > > > and watch a Japanese game, you won't see any kind > > > of silly moves (assuming they are not silly to a > > > Japanese observer). > > > > That's not true. The Chinese player (who has never > > heard of Japanese rules) will be confused by the > > "foolish" pass moves when the Japanese player > > refuses to fill dame. If he insists on his own > > limited perspective (like Japanese players often > > do) he will consider the Japanese player to be > > stupid for not taking those free points. > > > > I would point out that in Chinese, every stone on > > the board is territory, even if you define that > > to be "wrong." > > > > But it's not just the playing phase, it's the scoring > > phase where things are reckoned differently. This > > is where there is no right or wrong point of view, but > > you must pick one and agree on it. > > > > > > > > > The idea is: when there is > > > territory to be won, win it. That's the best you > > > can do by any ruleset. For strong players, the > > > ruleset does not make much difference (some minor > > > differences exist even between different Japanese > > > rulesets.) When there is no more territory to be > > > won, the game is finished, but that not easy to > > > understand for weak players. That's why the Chinese > > > ruleset indulgently ignores unne
RE: [computer-go] Interesting problem
People who play by Japanese rules fill the dame before passing and scoring. Professional game records leave those moves out since they are irrelevant, but if you go to a club and watch people playing, they usually fill the dame before passing. Sometimes you will see a verbal agreement that the game is over and both players will fill dame at the same time. A strong chinese player using chinese rules will pick up a point or two during the dame filling stage when playing a strong japanese player. The Chiense player will choose earlier moves that gain a later dame point that the japanese player will think have no benefit over other moves. Japanese rules has the minor advantage that they penalize plays into your opponent's terrioty that are not answered, so it is more risky in Japanese rules to make an unsound invasion at the end of the game. Chinese rules tolerates making these invasions and making safety moves to ensure aen enemy group is dead, or even taken off the board. This is why computer competitions like chinese rules. Japanese rules make the game more difficult. Chinese rules have the advantage of being more elegant. Disputes over group status can be played out. Japanese rules require agreement on group status, and there are rare situations that are very difficult to resolve. Most of the world plays be Japanese rules, so any commercial program must implement Japanese rules. David > -Original Message- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of Don Dailey > Sent: Sunday, December 31, 2006 7:08 AM > To: computer-go > Subject: Re: [computer-go] Interesting problem > > > On Sun, 2006-12-31 at 13:00 +, Jacques Basaldúa wrote: > > I don't agree on that. If you are used to Chinese > > and watch a Japanese game, you won't see any kind > > of silly moves (assuming they are not silly to a > > Japanese observer). > > That's not true. The Chinese player (who has never > heard of Japanese rules) will be confused by the > "foolish" pass moves when the Japanese player > refuses to fill dame. If he insists on his own > limited perspective (like Japanese players often > do) he will consider the Japanese player to be > stupid for not taking those free points. > > I would point out that in Chinese, every stone on > the board is territory, even if you define that > to be "wrong." > > But it's not just the playing phase, it's the scoring > phase where things are reckoned differently. This > is where there is no right or wrong point of view, but > you must pick one and agree on it. > > > > > The idea is: when there is > > territory to be won, win it. That's the best you > > can do by any ruleset. For strong players, the > > ruleset does not make much difference (some minor > > differences exist even between different Japanese > > rulesets.) When there is no more territory to be > > won, the game is finished, but that not easy to > > understand for weak players. That's why the Chinese > > ruleset indulgently ignores unnecessary moves. Moves > > made when the game is finished don't win anything > > and don't mean anything (neither by Chinese rules). > > They are objective wrong moves. > > Yes, from Japanese eyes they are, but from Chinese eyes > they are neutral. > > I could just as easily say that if these moves do > nothing, why are they actually penalized in Japanese? > I could say that Chinese is far more logical in that respect > and that Japanese rules are "wrong" because they > unfairly penalize you for moves that don't hurt your > position in any way. > > But I'm not going to say that, because I accept that > Japanese rules is a perfectly legitimate way to play > the game. It would really be silly of me to claim > they are "wrong" because they are indulgent about > not punishing PASS moves when there is territory to > be gained. > > > Imagine a chess > > game where you give mate, then you capture the > > king and, after that, you still move your pawns. > > If a human does that, its offensive for the > > opponent. If a computer does that, its just wrong. > > The chess analogy is wrong. The rules define checkmate > as the LAST MOVE. Game is over. There is nothing left > to resolve. > > Perhaps a better analogy is playing out a lost game in > Chess. Perhaps it's Q vs K and the losing side "makes" > the opponent checkmate him. Even in this case the > losing side can hope for an accidental stalemate, which > can happen with naive weaker players who are playing too > fast. > > Although some players may express annoyance
Re: [computer-go] Interesting problem
On Sun, 2006-12-31 at 13:00 +, Jacques Basaldúa wrote: > I don't agree on that. If you are used to Chinese > and watch a Japanese game, you won't see any kind > of silly moves (assuming they are not silly to a > Japanese observer). That's not true. The Chinese player (who has never heard of Japanese rules) will be confused by the "foolish" pass moves when the Japanese player refuses to fill dame. If he insists on his own limited perspective (like Japanese players often do) he will consider the Japanese player to be stupid for not taking those free points. I would point out that in Chinese, every stone on the board is territory, even if you define that to be "wrong." But it's not just the playing phase, it's the scoring phase where things are reckoned differently. This is where there is no right or wrong point of view, but you must pick one and agree on it. > The idea is: when there is > territory to be won, win it. That's the best you > can do by any ruleset. For strong players, the > ruleset does not make much difference (some minor > differences exist even between different Japanese > rulesets.) When there is no more territory to be > won, the game is finished, but that not easy to > understand for weak players. That's why the Chinese > ruleset indulgently ignores unnecessary moves. Moves > made when the game is finished don't win anything > and don't mean anything (neither by Chinese rules). > They are objective wrong moves. Yes, from Japanese eyes they are, but from Chinese eyes they are neutral. I could just as easily say that if these moves do nothing, why are they actually penalized in Japanese? I could say that Chinese is far more logical in that respect and that Japanese rules are "wrong" because they unfairly penalize you for moves that don't hurt your position in any way. But I'm not going to say that, because I accept that Japanese rules is a perfectly legitimate way to play the game. It would really be silly of me to claim they are "wrong" because they are indulgent about not punishing PASS moves when there is territory to be gained. > Imagine a chess > game where you give mate, then you capture the > king and, after that, you still move your pawns. > If a human does that, its offensive for the > opponent. If a computer does that, its just wrong. The chess analogy is wrong. The rules define checkmate as the LAST MOVE. Game is over. There is nothing left to resolve. Perhaps a better analogy is playing out a lost game in Chess. Perhaps it's Q vs K and the losing side "makes" the opponent checkmate him. Even in this case the losing side can hope for an accidental stalemate, which can happen with naive weaker players who are playing too fast. Although some players may express annoyance when a weaker player continues to play out a lost game, it's generally understood that it's your right to play out a game as far as you want. To me, it's far more rude to "pressure" a weaker player to resign by heavy sighs, rolling your eyes and demeaning him. The fact of the matter is that in Chinese, every stone is territory and in Japanese stones are just a kind of infrastructure around the territory. Although I respect both points of view, I don't like Japanese rules even though I understand them. One big beef I have with them is that it requires you to track all the captured stones. The board state isn't just the board, it's the board and 2 cups full of stones. Another beef is that if you move into your own territory, you get "double penalized." Notice how I am coloring my point of view by Chinese standards?If you have a good move in Chinese but you instead move into your own territory, you are punished in a natural way. Japanese is more heavy handed, you get the natural consequences of not playing the best move, but then you get slapped again for "reducing" your territory. From a Chinese players point of view, this is "wrong" because it's YOUR TERRITORY to being with! How can I be penalized for moving into my own territorydu! Please pardon me - for just a moment I was being narrow minded thinking Chinese was the only viewpoint. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
Don Dailey wrote: >Your odds of finding a "winning move against a pro >player" is different from finding one of the "best >move(s)" in the position, ... I agree. I was oversimplifying. It would be more appropriate to say: Except, probably for the first moves (as you point correctly, where the number of playable moves is higher), the average ratio between pro-level-playable-moves and legal-moves is *maybe* 1/100. There is sometimes only one (pro level) playable move and sometimes two or three in, perhaps, 200 legal moves (decreasing while the board fills). My numbers should only be taken as an orientation of how small the probability is. Of course, this does not pretend to be an argument against MC. It is an answer to a post who related this with infinite shift of Elo ratings against players who "always loose". As I understood, Aloril, used this to state that a random player does not necessarily loose always, even against a pro. I have been reading with much interest all that has been posted on MC in the last months and it sure is a promising way. I still have the doubt if it could have some kind of ceiling in cases where the more you approach the correct move the more you loose, as in ladders. (see my post "MC/UCT question" 6, December) On the Chinese/Japanese question: >This cuts both ways. Wouldn't it be pretty silly if >I was watching a Japanese game and continuously >criticized certain moves as "wrong" based on Chinese >standards? I don't agree on that. If you are used to Chinese and watch a Japanese game, you won't see any kind of silly moves (assuming they are not silly to a Japanese observer). The idea is: when there is territory to be won, win it. That's the best you can do by any ruleset. For strong players, the ruleset does not make much difference (some minor differences exist even between different Japanese rulesets.) When there is no more territory to be won, the game is finished, but that not easy to understand for weak players. That's why the Chinese ruleset indulgently ignores unnecessary moves. Moves made when the game is finished don't win anything and don't mean anything (neither by Chinese rules). They are objective wrong moves. Imagine a chess game where you give mate, then you capture the king and, after that, you still move your pawns. If a human does that, its offensive for the opponent. If a computer does that, its just wrong. Jacques. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
On Sat, 2006-12-30 at 13:52 +, Jacques Basaldúa wrote: > Aloril wrote: > > > Actually given *enough* games "fully random including > > eye filling and passing moves" will win against a pro player. > > That is "true", at least as it is true that a monkey would > write Hamlet typing at random long enough. > > That probability is in the range of 1 to (x·100)^(y·100) > where x and y are > 1. x represents the number of available > moves in hundreds (more than 100 typically) and y represents > the number of consecutive moves the random player has to > "guess". (The whole game and, therefore, also more than > hundred.) This number is below the probability of breaking > any type of cryptographic system (private key or public key) > by a fluke in the first guess. For practical reasons that > number is called zero. ;-) Your odds of finding a "winning move against a pro player" is different from finding one of the "best move(s)" in the position, you can get away with making some non-optimal moves, otherwise there would be no need for pro ranks. If a 4 dan pro needs 2 stones to get equal chances against a 6 dan pro, then there is a bit of wiggle room. It's easier to think in terms of a random player beating the ultimate player.I have no idea how many moves are playable in the ultimate sense from the opening position - but I can imagine that even surviving after playing 4 or 5 random moves will be most unlikely. But I imagine that the ultimate player can take a few random plays and still beat a strong pro, after all a strong pro can give 3 or 4 stones away for free against a weaker pro and still win! But a few randomly placed stones is better than nothing at all! Imagine if those stones were not placed completely random but had just a little bit of common sense added to them. The odds can be improved significantly with minor improvements, such as biases against certain moves. "Don't move in the corner" for the first N moves, for instance, will increase your odds of playing a complete game perfectly by orders of magnitude!But still not enough to make it likely in this universes lifetime. Many searching algorithms are based on this principle. Simulated annealing plays the odds like this with minor biases added to random searches which can actually produce reasonable moves a high percentage of the time. One of the first MC style programs used simulated annealing. All these types of algorithms are based on the principle that a random search given even the most minor help improves enormously, and random search even by itself is a reasonable way to explore a very large space and get useful information very quickly. Many people equate "random" search with "meaningless" search and I believe it is why there is much resistance to the idea. But random search can be directed quite nicely. A very early MC program of mine tried to "evolve" a strategy using a random search (PBIL) which is similar to simulated annealing and genetic algorithms.It was quite effective at this strategy, however there were serious limitations to the way I represented knowledge of the game, which put a low ceiling on what it could achieve. I think this is a very promising approach except that it's almost impossible to represent a strategy in a way that is compact, specific enough and fast enough to lend itself to a huge number of simulations. The way I represented a playing strategy was as a single list of moves, very much like Goggle, a go program that uses simulated annealing to find a move.What was being optimized was the ordering of the move list and the strategy was to play the first legal move in the list.This plays reasonable looking moves but is a very limited strategy. The ideal kind of knowledge is admissibly correct knowledge. Unfortunately, it's hard to come by in GO. Even the 1 point eye avoidance rule that seems so safe is not admissible. The rule, "don't move to the corners" for the first N moves is another one of these rules. Can it be proved that this is NEVER the best move in the first N plays of the game? Of course I expect N to be a reasonable number. I think it's clear that a corner move is not the best FIRST move, but I doubt this can actually be proved!Or maybe it can? If it can actually be taken advantage of or shown that a pass move is better than a corner move, that might be close to a proof, close enough that I would accept it. Of course I am already convinced it is not the best first move. Or can it at least be proved that in a perfectly played game, a corner move would never happen in the first N moves where N is a reasonable number? I could see how positions might be constructed early in the game where it IS the best move, but could it ever be the best move in a perfectly played game?Of course I don't expect an answer. The point is that admissible heuristics would go a LONG way in a random search - just a few simple ones us
Re: [computer-go] Interesting problem
Aloril wrote: Actually given *enough* games "fully random including eye filling and passing moves" will win against a pro player. That is "true", at least as it is true that a monkey would write Hamlet typing at random long enough. That probability is in the range of 1 to (x·100)^(y·100) where x and y are > 1. x represents the number of available moves in hundreds (more than 100 typically) and y represents the number of consecutive moves the random player has to "guess". (The whole game and, therefore, also more than hundred.) This number is below the probability of breaking any type of cryptographic system (private key or public key) by a fluke in the first guess. For practical reasons that number is called zero. ;-) The problem of "towards infinite shift" of Elo rating of programs vs random is a different one. It is caused because the probability formula never returns zero (except for infinite difference), but it does return very small values (not as small as those of my previous argument). One possible solution would be to force this value to zero (statisticians do that for problems related with goodness of fit) when the number of expected wins is below 5 in n. Where n a reasonable guess for the number of games the bot can play. Using this, there would be a maximum Elo difference above which nothing should be computed or, even better, the game should not be played at all. But there is, of course, the possibility of a win due to an Internet fail, a bug, the operator resigned to power off the computer .. and the this probability is equal for both. So probably doing nothing is also an wise option. ;-) Jacques. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
Is there a reason why we need to decide, in advance, which of these many candidates should be the anchorman? If we set up a whole swathe of them, surely a week of random even games answers many of these questions and gets us well on our way to a stable basis for a 19x19 competition? Maybe after the first 100 games between each pairing we can even play at random handicaps. I realise that there are questions that even this will not answer, but at least we will have numbers to argue over. I suggest: a) everyone who has knows of a reasonable contender for the role posts a URL and details of how to set it up. b) those of us who have access to a machine that can reasonably run a go player for a week offer to host / run one of these c) once the resource constraints become clear it may be possible to host more than one player per machine. I'm more than happy to volunteer a machine week for such a purpose, and a couple of hours to set a player up. Unfortunately my own computer-go player will probably not be robustly playable in time. cheers stuart ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
Le vendredi 29 décembre 2006 10:58, Aloril a écrit : > On Thu, 2006-12-28 at 11:53 +0100, alain Baeckeroot wrote: > > Le jeudi 28 décembre 2006 03:34, Don Dailey a écrit : > > > I'm having an interesting problem - my hope is to set > > > a random legal move making player (who doesn't fill > > > 1 point eyes) at ELO zero. > > Hmm maybe i misunderstand. It seems to me that a random player > > cannot have a fixed rating (except -infinity) as it will lose > > all his games against non random player. > > First thing: Don Dailey is talking about equivalent player to > PythonBrown in 9x9 CGOS, not Random. Oh , this might change by 200 ELO ! Pythonbrown's rank depends mainly on the number of weak program between it and the anchor. > > > Or if it is set at zero ELO all other non random bot will slowly > > drift toward +infinity. > > Actually given *enough* games "fully random including eye filling and > passing moves" will win against a pro player. The *enough* is the problem. Considering the sun will burn the earth in less than 10 billion year, it is statistically impossible that a random player will win a strong dan player before the earth disapear. Paraphrasing Linux coding style: "An infinite number of monkey typing in GNU Emacs, would never make a good go program" > I did find one game where > pseudorandom player (==Random at CGOS) wins against GNU Go on 5x5 board. > Haven't found one in 7x7 though. http://homepages.cwi.nl/~tromp/go/legal.html 5x5 is 414 billion possible games, 7x7 is 8.3 x 10^22 19x19 is 2 x 10^170 Dont waste your time looking for random win on 7X7, it will ~never happen. I won't discuss this topic anymore, it already happened 1 year ago at the beginning of cgos-9x9. Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
On Thu, 2006-12-28 at 11:53 +0100, alain Baeckeroot wrote: > Le jeudi 28 décembre 2006 03:34, Don Dailey a écrit : > > I'm having an interesting problem - my hope is to set > > a random legal move making player (who doesn't fill > > 1 point eyes) at ELO zero. > Hmm maybe i misunderstand. It seems to me that a random player > cannot have a fixed rating (except -infinity) as it will lose > all his games against non random player. First thing: Don Dailey is talking about equivalent player to PythonBrown in 9x9 CGOS, not Random. However, for remainder of mail I'm talking mainly about actual random player. I think it has finite, albeit negative rank at 9x9 CGOS. Random statistics against PythonBrown: Result: 348/2511 Win: 13.9% It manages enough often to pass in position where weak opponent has passed and its winning ;-) On 19x19 I think win % would be significantly lower. > Or if it is set at zero ELO all other non random bot will slowly > drift toward +infinity. Actually given *enough* games "fully random including eye filling and passing moves" will win against a pro player. I did find one game where pseudorandom player (==Random at CGOS) wins against GNU Go on 5x5 board. Haven't found one in 7x7 though. I don't know about if this holds true for "non eye filling random player", sometimes it's impossible for it to randomly to play best move ;-) -- Aloril <[EMAIL PROTECTED]> ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
Le jeudi 28 décembre 2006 18:47, Don Dailey a écrit : > Yes, > > Someone mentioned random as being infinitely weak but there is no > such thing.Resigning on the first move is as weak as you can > get. > > The random player isn't really random, it doesn't fill it's eyes. > There are strategies to play MUCH worse than random as you > point out. Ok maybe one can build worst than random program, they will converge faster toward -infinity :) > > I don't believe the random player is particularly easy to take > advantage of. Of course it's very easy to beat, but it has > no special "quirks" compared to more deterministic algorithms. On CGOS 9X9 "GNUGo-1.2" is nearly 1000 ELO stronger than "Random", and on kgs gnugo1pt2 is ranked 22k, this is incredibly weak. 10-games-beginner is stronger than this ! I think it would be totally useless to try to use such a randombot for reference. The only use of random bot is for bug squashing in other engines, and test for time limit. Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
Yes, Someone mentioned random as being infinitely weak but there is no such thing.Resigning on the first move is as weak as you can get. The random player isn't really random, it doesn't fill it's eyes. There are strategies to play MUCH worse than random as you point out. I don't believe the random player is particularly easy to take advantage of. Of course it's very easy to beat, but it has no special "quirks" compared to more deterministic algorithms. That's part of the reason I feel it's a good baseline. - Don On Thu, 2006-12-28 at 18:19 +0100, [EMAIL PROTECTED] wrote: > I don't think it is possible to determine the weakest possible level. > One could make a program that plays much worse than random. It would > for instance rather fill its eyes than pass. I don't think any > handicap will be enough to allow such a player to win. Especially when > the (human) opponent knows the anti-tactics that were built into the > program. > > Dave > > - Oorspronkelijk bericht - > > Van: Don Dailey <[EMAIL PROTECTED]> > > Datum: donderdag, december 28, 2006 4:23 pm > > Onderwerp: Re: [computer-go] Interesting problem > > > > > > > On Wed, 2006-12-27 at 22:53 -0500, Don Dailey wrote: > > > It turns out that I did not turn off all of the stuff > > > that strengthened the random player - so hopefully I > > > have much weaker players now. > > > > > > (There was a bug that made the program too strong :-) > > > > > > - Don > > > > Addendum: > > > > However, there still is a large gap between purely random > > and 1 simulation MC. Well over 100 ELO points. > > > > - 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 mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
I don't think it is possible to determine the weakest possible level. One could make a program that plays much worse than random. It would for instance rather fill its eyes than pass. I don't think any handicap will be enough to allow such a player to win. Especially when the (human) opponent knows the anti-tactics that were built into the program. Dave - Oorspronkelijk bericht - Van: Don Dailey <[EMAIL PROTECTED]> Datum: donderdag, december 28, 2006 4:23 pm Onderwerp: Re: [computer-go] Interesting problem > > > On Wed, 2006-12-27 at 22:53 -0500, Don Dailey wrote: > > It turns out that I did not turn off all of the stuff > > that strengthened the random player - so hopefully I > > have much weaker players now. > > > > (There was a bug that made the program too strong :-) > > > > - Don > > Addendum: > > However, there still is a large gap between purely random > and 1 simulation MC. Well over 100 ELO points. > > - 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] Interesting problem
On Wed, 2006-12-27 at 22:53 -0500, Don Dailey wrote: > It turns out that I did not turn off all of the stuff > that strengthened the random player - so hopefully I > have much weaker players now. > > (There was a bug that made the program too strong :-) > > - Don Addendum: However, there still is a large gap between purely random and 1 simulation MC. Well over 100 ELO points. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
anyone who plays by the rules qualifies as 30kyu. s. __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
Le jeudi 28 décembre 2006 03:34, Don Dailey a écrit : > I'm having an interesting problem - my hope is to set > a random legal move making player (who doesn't fill > 1 point eyes) at ELO zero. Hmm maybe i misunderstand. It seems to me that a random player cannot have a fixed rating (except -infinity) as it will lose all his games against non random player. Or if it is set at zero ELO all other non random bot will slowly drift toward +infinity. > > I feel this would define a nice standard that is > easy to reproduce and verify experimentally and > at least would be a known quantity even 100 years > from now. Any GPL go engine fulfill the above requirements :) Aloril did gtp patch for old gnugo, so it is possible to use gnugo-1.2 and 2.0 as weak players, and gnugo 3.7.10 at default level for stronger anchor (lower level are very poor in reading). Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
This is an interesting problem. It seems to me that the reality is that when you are talking about non-ideal play, ranking systems aren't linear. Program A could beat B which could beat C which could beat A. How would you rank those? Clearly there is going to have to be some degree of arbitrary selection. I propose convenience as the best reason for picking one anchor over another. I think a completely random player is the only other choice from a theoretically perfect player that doesn't have arbitrariness. But, by defining players relative to that anchor, we would really be measuring how effectively a program exploits a weak player rather than how good the program is. It is my opinion that it is more important to have a relative ranking system than an absolute system. - Nick On 12/28/06, Aloril <[EMAIL PROTECTED]> wrote: On Wed, 2006-12-27 at 21:34 -0500, Don Dailey wrote: > I'm having an interesting problem - my hope is to set > a random legal move making player (who doesn't fill > 1 point eyes) at ELO zero. > > I feel this would define a nice standard that is > easy to reproduce and verify experimentally and > at least would be a known quantity even 100 years > from now. > > But I'm having a difficult time creating players > who are slightly better than this at 19x19. I need > incrementally better and better players. I suspect this is quite hard problem. On 9x9 we have some of this and I suspect even there "do not fill eyes random" (PythonBrown) has not yet settled (maybe 100-200 ELO overrated). Probably too few weak players ;-) On 19x19 I think problem is much harder and required amount of intermediate players is much bigger. I'm of course interested in hearing your experimentation results. Maybe I'm wrong and it is actually feasible. My vague recollection was that random player is maybe 200 kuy, "do not fill eyes" adds 60 stones, atari detection adds about 20-30 stones, idiotbot is maybe 100 kuy, weakbot50k maybe 50 kuy. However differences between computers tend to be much bigger than when they play against humans! For example GNU Go 2.0 can give Liberty 1.0 easily 9 stones and win more than 50% of games (based on few ha9 test games), but at KGS they are rated at 10k and 14k. Even WeakBot50k is rated at 20k while latest GNU Go rated at 6k can give it numerous handicap stones (much more than 14 stones, I think it's more than 40 stones). Here is my proposal for anchor player: Use GNU Go 3.7.10 (or any enough recent with super-ko support) at level 0 and use well defined randomization on top of moves it returns. Ie. ask all_move_values (lists only moves that gnugo considers positive) and add remaining moves and then apply slight randomization so that it still plays close to original strength but is much more unpredictable than GNU Go. Example program (by blubb and me): http://londerings.cvs.sourceforge.net/londerings/go/gtpTuner/ Reasons: - reasonably strong, no need for huge amount of intermediate players - source code available - well known entity - with some randomization should be unpredictable I suspect that GNU Go without randomization is too predictable. This is very clearly case on 9x9 board and possibly on 19x19 too. -- Aloril <[EMAIL PROTECTED]> ___ 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] Interesting problem
On Wed, 2006-12-27 at 21:34 -0500, Don Dailey wrote: > I'm having an interesting problem - my hope is to set > a random legal move making player (who doesn't fill > 1 point eyes) at ELO zero. > > I feel this would define a nice standard that is > easy to reproduce and verify experimentally and > at least would be a known quantity even 100 years > from now. > > But I'm having a difficult time creating players > who are slightly better than this at 19x19. I need > incrementally better and better players. I suspect this is quite hard problem. On 9x9 we have some of this and I suspect even there "do not fill eyes random" (PythonBrown) has not yet settled (maybe 100-200 ELO overrated). Probably too few weak players ;-) On 19x19 I think problem is much harder and required amount of intermediate players is much bigger. I'm of course interested in hearing your experimentation results. Maybe I'm wrong and it is actually feasible. My vague recollection was that random player is maybe 200 kuy, "do not fill eyes" adds 60 stones, atari detection adds about 20-30 stones, idiotbot is maybe 100 kuy, weakbot50k maybe 50 kuy. However differences between computers tend to be much bigger than when they play against humans! For example GNU Go 2.0 can give Liberty 1.0 easily 9 stones and win more than 50% of games (based on few ha9 test games), but at KGS they are rated at 10k and 14k. Even WeakBot50k is rated at 20k while latest GNU Go rated at 6k can give it numerous handicap stones (much more than 14 stones, I think it's more than 40 stones). Here is my proposal for anchor player: Use GNU Go 3.7.10 (or any enough recent with super-ko support) at level 0 and use well defined randomization on top of moves it returns. Ie. ask all_move_values (lists only moves that gnugo considers positive) and add remaining moves and then apply slight randomization so that it still plays close to original strength but is much more unpredictable than GNU Go. Example program (by blubb and me): http://londerings.cvs.sourceforge.net/londerings/go/gtpTuner/ Reasons: - reasonably strong, no need for huge amount of intermediate players - source code available - well known entity - with some randomization should be unpredictable I suspect that GNU Go without randomization is too predictable. This is very clearly case on 9x9 board and possibly on 19x19 too. -- Aloril <[EMAIL PROTECTED]> ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
I haven't really paid much attention to the previous few emails, but if it is an effort at making a weak player (as the thread started out), I shouldn't have to. Why not just randomly chose (with a programmable distribution) between making a move based on a simulation and a completely random move? This will give you adjustable play quality somewhere between the two. On 12/28/06, [EMAIL PROTECTED] <[EMAIL PROTECTED]> wrote: A more detailed version. 1| 208 110 63 89 93 104 106 98 117 139 117 98 106 104 93 89 63 110 208 2| 110 18868 12 17 17 22 39 22 17 17 12 868 18 110 3| 638...266454662 ...8 63 4| 896..2555444555 2..6 89 5| 938.276 10 12 12 18 12 12 106 72.8 93 6| 104 12256 11 14 16 17 23 17 16 14 11 652 12 104 7| 106 1765 10 14 27 22 21 35 21 22 27 14 1056 17 106 8| 98 1765 12 16 22 39 29 35 29 39 22 16 1256 17 98 9| 117 2244 12 17 21 29 35 37 35 29 21 17 1244 22 117 10| 139 3954 18 23 35 35 37 178 37 29 35 23 1845 39 139 11| 117 2244 12 17 21 29 35 37 35 29 21 17 1244 22 117 12| 98 1765 12 16 22 39 29 35 29 39 22 16 1256 17 98 13| 106 1765 10 14 27 22 21 35 21 22 27 14 1056 17 106 14| 104 12256 11 14 16 17 23 17 16 14 11 652 12 104 15| 938.276 10 12 12 18 12 12 106 72.8 93 16| 896..2555444555 2..6 89 17| 638...266454662 ...8 63 18| 110 18868 12 17 17 22 39 22 17 17 12 868 18 110 19| 208 110 63 89 93 104 106 98 117 139 117 98 106 104 93 89 63 110 208 Dave Hillis -Original Message- From: [EMAIL PROTECTED] To: [EMAIL PROTECTED] Cc: computer-go@computer-go.org Sent: Wed, 27 Dec 2006 11:58 PM Subject: Re: [computer-go] Interesting problem Thanks Dave, - Don On Wed, 2006-12-27 at 23:50 -0500, [EMAIL PROTECTED] wrote: > File attached. And also inline below Dave Hillis antminder on KGS Check Out the new free AIM(R) Mail -- 2 GB of 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/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
A more detailed version. 1| 208 110 63 89 93 104 106 98 117 139 117 98 106 104 93 89 63 110 208 2| 110 18868 12 17 17 22 39 22 17 17 12 868 18 110 3| 638...266454662 ...8 63 4| 896..2555444555 2..6 89 5| 938.276 10 12 12 18 12 12 106 72.8 93 6| 104 12256 11 14 16 17 23 17 16 14 11 652 12 104 7| 106 1765 10 14 27 22 21 35 21 22 27 14 1056 17 106 8| 98 1765 12 16 22 39 29 35 29 39 22 16 1256 17 98 9| 117 2244 12 17 21 29 35 37 35 29 21 17 1244 22 117 10| 139 3954 18 23 35 35 37 178 37 29 35 23 1845 39 139 11| 117 2244 12 17 21 29 35 37 35 29 21 17 1244 22 117 12| 98 1765 12 16 22 39 29 35 29 39 22 16 1256 17 98 13| 106 1765 10 14 27 22 21 35 21 22 27 14 1056 17 106 14| 104 12256 11 14 16 17 23 17 16 14 11 652 12 104 15| 938.276 10 12 12 18 12 12 106 72.8 93 16| 896..2555444555 2..6 89 17| 638...266454662 ...8 63 18| 110 18868 12 17 17 22 39 22 17 17 12 868 18 110 19| 208 110 63 89 93 104 106 98 117 139 117 98 106 104 93 89 63 110 208 Dave Hillis -Original Message- From: [EMAIL PROTECTED] To: [EMAIL PROTECTED] Cc: computer-go@computer-go.org Sent: Wed, 27 Dec 2006 11:58 PM Subject: Re: [computer-go] Interesting problem Thanks Dave, - Don On Wed, 2006-12-27 at 23:50 -0500, [EMAIL PROTECTED] wrote: > File attached. And also inline below Dave Hillis antminder on KGS Check Out the new free AIM(R) Mail -- 2 GB of storage and industry-leading spam and email virus protection. 1| 208 110 63 89 93 104 106 98 117 139 117 98 106 104 93 89 63 110 208 2| 110 18868 12 17 17 22 39 22 17 17 12 868 18 110 3| 638...266454662 ...8 63 4| 896..2555444555 2..6 89 5| 938.276 10 12 12 18 12 12 106 72.8 93 6| 104 12256 11 14 16 17 23 17 16 14 11 652 12 104 7| 106 1765 10 14 27 22 21 35 21 22 27 14 1056 17 106 8| 98 1765 12 16 22 39 29 35 29 39 22 16 1256 17 98 9| 117 2244 12 17 21 29 35 37 35 29 21 17 1244 22 117 10| 139 3954 18 23 35 35 37 178 37 29 35 23 1845 39 139 11| 117 2244 12 17 21 29 35 37 35 29 21 17 1244 22 117 12| 98 1765 12 16 22 39 29 35 29 39 22 16 1256 17 98 13| 106 1765 10 14 27 22 21 35 21 22 27 14 1056 17 106 14| 104 12256 11 14 16 17 23 17 16 14 11 652 12 104 15| 938.276 10 12 12 18 12 12 106 72.8 93 16| 896..2555444555 2..6 89 17| 638...266454662 ...8 63 18| 110 18868 12 17 17 22 39 22 17 17 12 868 18 110 19| 208 110 63 89 93 104 106 98 117 139 117 98 106 104 93 89 63 110 208 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
Thanks Dave, - Don On Wed, 2006-12-27 at 23:50 -0500, [EMAIL PROTECTED] wrote: > File attached. And also inline below Dave Hillis antminder on KGS ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
File attached. And also inline below Dave Hillis antminder on KGS 1| 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2| 21 18868 12 17 17 21 21 21 17 17 12 868 18 21 3| 218...266454662 ...8 21 4| 216..2555444555 2..6 21 5| 218.276 10 12 12 18 12 12 106 72.8 21 6| 21 12256 11 14 16 17 21 17 16 14 11 652 12 21 7| 21 1765 10 14 21 21 21 21 21 21 21 14 1056 17 21 8| 21 1765 12 16 21 21 21 21 21 21 21 16 1256 17 21 9| 21 2144 12 17 21 21 21 21 21 21 21 17 1244 21 21 10| 21 2154 18 21 21 21 21 21 21 21 21 21 1845 21 21 11| 21 2144 12 17 21 21 21 21 21 21 21 17 1244 21 21 12| 21 1765 12 16 21 21 21 21 21 21 21 16 1256 17 21 13| 21 1765 10 14 21 21 21 21 21 21 21 14 1056 17 21 14| 21 12256 11 14 16 17 21 17 16 14 11 652 12 21 15| 218.276 10 12 12 18 12 12 106 72.8 21 16| 216..2555444555 2..6 21 17| 218...266454662 ...8 21 18| 21 18868 12 17 17 21 21 21 17 17 12 868 18 21 19| 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 -Original Message- From: [EMAIL PROTECTED] To: computer-go@computer-go.org Sent: Wed, 27 Dec 2006 11:43 PM Subject: Re: [computer-go] Interesting problem Can you send me an attachment with the 19x19 data in a text file? I will try a version for the 19x19 games and see what happens. - Don On Wed, 2006-12-27 at 23:35 -0500, [EMAIL PROTECTED] wrote: > I'll post a 19x19 version if anyone is interested, but the lines will > wrap around... ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Check Out the new free AIM(R) Mail -- 2 GB of storage and industry-leading spam and email virus protection. 1| 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2| 21 18868 12 17 17 21 21 21 17 17 12 868 18 21 3| 218...266454662 ...8 21 4| 216..2555444555 2..6 21 5| 218.276 10 12 12 18 12 12 106 72.8 21 6| 21 12256 11 14 16 17 21 17 16 14 11 652 12 21 7| 21 1765 10 14 21 21 21 21 21 21 21 14 1056 17 21 8| 21 1765 12 16 21 21 21 21 21 21 21 16 1256 17 21 9| 21 2144 12 17 21 21 21 21 21 21 21 17 1244 21 21 10| 21 2154 18 21 21 21 21 21 21 21 21 21 1845 21 21 11| 21 2144 12 17 21 21 21 21 21 21 21 17 1244 21 21 12| 21 1765 12 16 21 21 21 21 21 21 21 16 1256 17 21 13| 21 1765 10 14 21 21 21 21 21 21 21 14 1056 17 21 14| 21 12256 11 14 16 17 21 17 16 14 11 652 12 21 15| 218.276 10 12 12 18 12 12 106 72.8 21 16| 216..2555444555 2..6 21 17| 218...266454662 ...8 21 18| 21 18868 12 17 17 21 21 21 17 17 12 868 18 21 19| 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
Can you send me an attachment with the 19x19 data in a text file? I will try a version for the 19x19 games and see what happens. - Don On Wed, 2006-12-27 at 23:35 -0500, [EMAIL PROTECTED] wrote: > I'll post a 19x19 version if anyone is interested, but the lines will > wrap around... ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Interesting problem
Here is a way to make a slightly smarter (and much prettier) random player. It is good for MC playout games too. For the first 20 or so plys, restrict the board spaces that can be filled to those commonly seen in human games. Below is a table for 9x9 games (trained from SGF files). It is best seen in proportional font. abcdefghi 1| 21 17 16 14 13 14 16 17 21 2| 17 1064346 10 17 3| 166.....6 16 4| 144.....4 14 5| 133.....3 13 6| 144.....4 14 7| 166.....6 16 8| 17 1064346 10 17 9| 21 17 16 14 13 14 16 17 21 The random player can place a stone on any of the innermost 5x5 spaces (those with a '.') at any ply, but it can't place a stone on e2 before the 3rd ply, or b1 before the 17th ply, or a1 before the 21st ply, etc. I'll post a 19x19 version if anyone is interested, but the lines will wrap around... Dave Hillis antminder on KGS -Original Message- From: [EMAIL PROTECTED] To: computer-go@computer-go.org Sent: Wed, 27 Dec 2006 10:53 PM Subject: Re: [computer-go] Interesting problem It turns out that I did not turn off all of the stuff that strengthened the random player - so hopefully I have much weaker players now. (There was a bug that made the program too strong :-) - Don On Wed, 2006-12-27 at 21:34 -0500, Don Dailey wrote: > I'm having an interesting problem - my hope is to set > a random legal move making player (who doesn't fill > 1 point eyes) at ELO zero. > > I feel this would define a nice standard that is > easy to reproduce and verify experimentally and > at least would be a known quantity even 100 years > from now. > > But I'm having a difficult time creating players > who are slightly better than this at 19x19. I need > incrementally better and better players. > > But even a monte carlo program that does 1 simulation > rarely loses, if my experiment is correct. > > The way it's coded has this effect conceptually: > > 1. Pick a random move R > > 2. Play a random game from R > > 3. If the game is a win, play R, otherwise > pick some other move chosen randomly. > > > It's interesting that this strategy is so much stronger > than completely random play, because it's very close > to random, or so I thought. > > I guess I have to make even this strategy more random. > > I never thought I would have trouble making a weak > player! > > > - 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/ Check Out the new free AIM(R) Mail -- 2 GB of 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] Interesting problem
It turns out that I did not turn off all of the stuff that strengthened the random player - so hopefully I have much weaker players now. (There was a bug that made the program too strong :-) - Don On Wed, 2006-12-27 at 21:34 -0500, Don Dailey wrote: > I'm having an interesting problem - my hope is to set > a random legal move making player (who doesn't fill > 1 point eyes) at ELO zero. > > I feel this would define a nice standard that is > easy to reproduce and verify experimentally and > at least would be a known quantity even 100 years > from now. > > But I'm having a difficult time creating players > who are slightly better than this at 19x19. I need > incrementally better and better players. > > But even a monte carlo program that does 1 simulation > rarely loses, if my experiment is correct. > > The way it's coded has this effect conceptually: > > 1. Pick a random move R > > 2. Play a random game from R > > 3. If the game is a win, play R, otherwise > pick some other move chosen randomly. > > > It's interesting that this strategy is so much stronger > than completely random play, because it's very close > to random, or so I thought. > > I guess I have to make even this strategy more random. > > I never thought I would have trouble making a weak > player! > > > - 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] Interesting problem
I'm having an interesting problem - my hope is to set a random legal move making player (who doesn't fill 1 point eyes) at ELO zero. I feel this would define a nice standard that is easy to reproduce and verify experimentally and at least would be a known quantity even 100 years from now. But I'm having a difficult time creating players who are slightly better than this at 19x19. I need incrementally better and better players. But even a monte carlo program that does 1 simulation rarely loses, if my experiment is correct. The way it's coded has this effect conceptually: 1. Pick a random move R 2. Play a random game from R 3. If the game is a win, play R, otherwise pick some other move chosen randomly. It's interesting that this strategy is so much stronger than completely random play, because it's very close to random, or so I thought. I guess I have to make even this strategy more random. I never thought I would have trouble making a weak player! - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
