Re: [Computer-go] 37x37 tournament on KGS
Any estimates how long the games will be? Even 21x21 games i've seen land clearly 'not very entertainig' domain. PP 2014-07-14 13:47 GMT+03:00 Ingo Althöfer 3-hirn-ver...@gmx.de: Hello guys, in the insane section of KGS a tournament on 37x37-boards will be played next week: http://www.gokgs.com/tournInfo.jsp?id=910 It would be interesting to see how bots perform there. Is someone willing to let his baby participate? Ingo. PS. Some years ago Zen participated in a 21x21-tournament on KGS and got an impressive second rank, if I remember correctly. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Skip-opening matchmode
My hypothesisi is that weird opening is not good or bad objectively, but works because it matches the playing style of moves to come. And small differences in opening do not matter much as it is the middle game that sets players a part. Difference in points between 3-4 and 10-4 as 1st move is probably way below one, and in any middle game fights difference between good and almost good moves is almost always more than 10 points. So good or bad opening is something that will be an issue once bots are playing pro's with 1-2 stone handicap. In chess I notices load of player buy and read opening books while at they skill level they hardly ever meet an opponent that will play a know opening past move six. And they lose and win games on dropped pieces on fine nuances of strategy. Similarly in go loads of people study joseki and lose games on not reading medium hard tsume-go situations correctly Petri 2014-06-04 15:51 GMT+03:00 Marc Landgraf mahrgel...@gmail.com: Hi, another idea crossed my mind lately. We see a lot of Bots play rather unconventional fuseki. Sadly it looks difficult to know, if those are actually a weakness or a strength of certain bots as our human judgement is not perfectly accurate here either. Thus I would love to see some games, in which the game is not started on an empty board, but on positions from professional games at around move 50 or 60. (the position of course should be not known to the bot before, so no preanalysis) For fairness reasons the players of course have to play it twice with alternating colors to prevent any potential advantage from the given board position. It would be interesting to see, if those Bot fuseki are actually playing into the bots strengths or if they are handicapping the bot, as they would do better with conventional openings, but are just unable to play them. Of course, the best would be, if it would be possible to somehow test it against humans of appropriate level, but I'm not sure, how this could be done. But even some sort of bot tourney with this mode would probably be interesting, even though it would not tell us, what I described above, but still the comparison to a normal tourney may tell something about strengths of certain bots. Same could be also done with endgame positions, but finding endgame positions that are challenging for the bots without being predecided on their level is probably rather difficult. ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] [O-T] Arimaa
I think both are too fat for truly analytical answer. But number of people working in computer go vs arimaa and number of people playing the two games indicates that Go would be the 1st to have computer beating top humans. Last time I logged in into arimaa server there were 11 people there. Drawing from such a shallow talent pool coming up with good ideas will take huge time. And I do not think Arimaa has chance. It lacks the violent nature chess nor is strategy as subtle as in Go - not to mention violent nature of Go which in my games usually happens the wrong way around. And most importantly if I want to play I can easily find someone to play chess with me, with bit of trouble i would find someone to play Go but for Arimaa... I would need to teach rules to someone. And in in these games competitive factor is huge, no competition id there is no one to compete with. PP ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Congratulations to Crazy Stone!
I am afraid can depend on rule set in use. I guess that under Japanese rules bent four is dead by definition. But having two of them does not really change anything as attacker can choose to fight then after each other. Petri 2013/12/10 Stefan Kaitschick stefan.kaitsch...@hamburg.de What happens with 2 bent fours? You should be able to save one of them. On Mon, Dec 9, 2013 at 10:01 PM, Hiroshi Yamashita y...@bd.mbn.or.jp wrote: In round 3 CrazyStone vs. Aya, probably Aya was expecting W to play L12 or N12, if B's corner was not 100% dead in the playouts. It seems a bug in Aya's playouts, but even if there was no bug, Aya could still reasonably expect W's L12 or N12 in the tree. Corner bent-four is not easy to handle Yesterday I turned off bent4 code for test, and forgot. Aya understands bent4 as seki. And at end of playout, bent4 shape is counted as dead. It works well, but when outside of bent4 is semeai, it does not work. Maybe when outside W lib is 5, W has to play N12? And thanks for nice look cross-table and report. I was bit surprised I got runners-up 7 times in a row this year. Regards, Hiroshi Yamashita ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] How many probes down the tree are necessary for a good bot?
3013 :) You are really ahead of us. I am KGS 4-5k and Manyfaces on two core 1,5 GHz will usually beat me. If I take 2 stone handicap I usually beat it. And I always use lopsided time controls, computer is not nearly impatient as I am, something like 30 min for me and 15 mins for it and 30 secs byo-yomi. I think bots are relatively stonger on non-handicap games. I can check how many sims it does for that but not 100 000 I would say. And 5k KGS is probably round 7k Finnish Go Association. So ratings are just as confusing in go as they are in chess servers f.ex (try to count on chess.com forums how many threads there are on subject 'how does this rating related to that rating') 2013/11/13 Darren Cook dar...@dcook.org How many probes/playouts are necessary for a 19x19 MCTS bot to beat a human amateur? Let's say 5k rating. (5 kyu will need defining a bit more: can we assume a KGS 5-kyu?) If light playouts (i.e. random, with just the most basic go knowledge to prevent filling your own eyes and ko checking to avoid infinite loops), I don't think there is any program that can beat a 5-kyu. So, my answer is lots and lots. At the other end, with heavy playouts some of the programs sound like they might be playing at that level with 100K playouts? (That is a very rough guess; I'm going by the Measuring program strength thread (Aug 3013) where people are getting 50% against GnuGo with 1K to 10K playouts.) Darren ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] On Semeai Detection
Thats why Gmail has filters and you can prevent gmail classifying as spam from specific senders. Petri 2013/10/9 Aja Huang ajahu...@gmail.com I didn't get Don's goodbye message as well, and I found Gmail filtered lots of emails from the list as spams. Aja 2013/10/9 Erik van der Werf erikvanderw...@gmail.com On Wed, Oct 9, 2013 at 8:54 PM, Ingo Althöfer 3-hirn-ver...@gmx.de wrote: Ps. Sorry for writing this as a new message. But for months already I do not receive the mails from the list, and an attempt to register from another mailing address failed. You're not the only one with email problems. Last month I didn't get Don's goodbye message (but saw it in the archive). Then later I didn't get the follow-up replies from Don and GCP... Most emails still seem to get through to me, but it's a pity that this list is becoming unreliable. Erik ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] is the list still working?
Whic can be prevented by gmail filters. My gmail says something like this mail has not been sent to spam because of filter you created. It says in in Finnish so all typos and bad grammar re from me not from google :) 2013/8/20 Alexander Kozlovsky alexander.kozlov...@gmail.com It works for me and for others: http://dvandva.org/pipermail/computer-go/2013-August/thread.html But GMail marks almost all computer-go messages as spam, so I need to check spam folder regularly. On Tue, Aug 20, 2013 at 6:08 AM, Ray Tayek rta...@ca.rr.com wrote: i have no posts since 7-22. thanks --- co-chair http://ocjug.org/ __**_ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/**mailman/listinfo/computer-gohttp://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Question on 0.5-wins
Bots also fill in their own territory removing very distant threats if they can afford to do it. Probably saves a loss in one game out 100 or so. Petri 2013/6/25 Stefan Kaitschick stefan.kaitsch...@hamburg.de I have never understood this bot behaviour, because once a position gets very close, every move becomes critical and that should depress the winrate atleast somewhat. The only explanation that I have for myself, is that while the lead is still comfortable, the bot will shun optimal moves if they require any kind of followup, because that looks less safe to the bot. Stefan ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Narrow wins
I can tell from experience that it certainly affects how bots when giving handicap. I dob't mfog comercially availalable version has it but my first feeling ehrn playing it was it does not. On 4 core laptop I pretty much lost 4 game out of five even but had no trouble with two stone handicap. Without synamic go bot try to get even in opening stage which is not really doable if handicap is correct. With Pachi - I assume due ti dynamic handicap - situation feels different. Measuring this effect is possible since bots can be easily adjusted in strength, but takes quite an effort still. Petri 2013/6/11 Darren Cook dar...@dcook.org The dynamic komi can be pretty large. I think I limit it to 30 points or so. If the komi were only a point or two you would see a lot of games won by 2.5 points or so. You can see from the results that the margins are often pretty large. Hello David, Interesting! Is the benefit just from more natural-looking play (from the point of view of humans), or is it actually increasing the win rate do you think? (And if so, just the win rate against humans, or against other programs too?) Darren ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Playing strenght versus respond time
I have only anecdotal evidence but still I am pretty sure of my observation: Faster I play faster my opponents respond - usually - regardless of time limits. So playing fast may well trigger bad moves by opponent. Petri 2013/6/4 Detlef Schmicker d...@physik.de I wonder if somebody tested the human playing strength versus bot respond time on kgs (with the same bot parameters otherwize). The reason I ask is: I want to test the scaling with playouts_per_move. Aya is so kind to offer its playouts per move and I wondered if the relatively bad scaling has to do with humans playing more serious if the bot takes more time? Detlef ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] MCTS + Neural Networks?
I would say you would loose too many simulations. Besides by using whatever power to increase simulations/second probably gives better results. Optimizing simulations is a dark art. I think there are several test to show thath if you make the simulation AI better the it may make your bot weaker, even with similar amount sims/move. Perhaps applying neural nets in tree search part to bias the search? Like Many Faces does with opening book. Petri 2013/5/2 Steven Clark steven.p.cl...@gmail.com Thanks for the link! Looks like a good paper -- I will read it more carefully shortly. Ignoring computational speed for a moment, is it a reasonable assumption that an algorithm that plays a NN-proposed tactical move 50% of the time, and a random move 50% of the time, should outperform an algorithm that plays a random move 100% of the time? So it's just a case of how many playouts do we lose by employing the NN (GPUs to the rescue?). For reference, I was using 25 input nodes, 25 output nodes, ~50 hidden nodes. I guess ultimately it comes down to make a bot and prove it :) -Steven On Wed, May 1, 2013 at 9:50 PM, George Dahl gd...@cs.toronto.edu wrote: I don't know if neural nets that predict moves have been helpful in any strong bots, but predicting expert moves with neural nets is certainly old news. See http://www.cs.utoronto.ca/~ilya/pubs/2008/go_paper.pdf There might be a place for artificial neural nets in a strong Go playing program, but it is an open question on how to incorporate neural nets well. Software like neurgo used a lot of expert features along with a neural net for global position evaluation and I tried (with very little success) to predict ownership of points on the board using a neural net. It is very hard to get neural nets to help a standard MCTS bot a lot because the neural net needs to be good at whatever it is supposed to be doing and still probably very fast to be useful. - George On Wed, May 1, 2013 at 9:42 PM, Steven Clark steven.p.cl...@gmail.comwrote: Hello all- Has anyone successfully used neural nets to help guide MC playouts? Has anyone used NN to learn patterns larger than 3x3? I'm working on a grad-school project, and discovered a few interesting things. After analyzing 10,000+ high-dan games from KGS, I find that more than 50% of the time, moves are played within a 5x5 window centered at the opponent's previous move (call this a tactical move, vs a strategic move). I used the FANN library to learn these 5x5 patterns, and found that the NN could predict tactical moves with ~27% accuracy (and with a 50% chance that the answer would be in the top 3 moves proposed by the NN). Is this old news? Are neural nets just too slow to be helpful to MC (reduce the playout rate too much?) Thoughts welcome. I will be up late finishing the report since it is due tomorrow ;) -Steven ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Weight of moves
They are random so very little information can be taken out. Also about local fights and grand level strategy: Since the MC program always do a full board search they - relatively speaking - are masters at using full board and connecting local fights to global ones. When I used my laptop to run ManyFacesOfGo I quite often lost on getting hammered in fights involving more than one area of the board. So MC makes a plan for the game, it just makes it implicitly and you cannot ask from it what it is. Petri 2013/4/1 Gabriel .Santos gabrielmsan...@gmail.com Thanks Aja. I was just wondering if it is possible to get more relevant information from the playouts and take advantage of that. =D. On Mon, Apr 1, 2013 at 3:58 PM, Aja Huang ajahu...@gmail.com wrote: 2013/4/1 Gabriel .Santos gabrielmsan...@gmail.com 3 - I guess you misunderstood my idea of think, or perhaps i just choose the wrong word. By the question that I raise, i mean, how anyone say that one state is better than another one if they have the same winrate in MC methods. How could the machine determine this ? If one of the states has more trials than the other, then the machine determines it is better with higher confidence. Aja ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] 50k-100k patterns
2012/8/10 Mark Boon tesujisoftw...@gmail.com It's hard to find people at age 20 or 30 to start to devote their life to something new. But it does happen on occasion. About 10 years ago there was someone like that at the Amsterdam Go Club. Strong chess player. Well over 30, unemployed I believe, who spent day and night studying Go and playing online. At the time I told him the same I have been saying here. But he didn't believe me. I think some 2 years later he was 1-dan, which was already very promising. But I think becoming 2-dan then took another two years. I don't know what happened after that, but I believe he got stuck at 2 or 3 dan. Mark Age has an impact for sure. In learning new skill there is myelin involved. Last time person gets bigger influx of myelin is around 30's. So up to taht learning skill should be quite doable. But doable does not mean easy. Proper training is hard and annoying. I do not know what is best way to train for go, but one thing dor sure it will not involve huge amount of playing go, obviously that will not hurt but that cannot be the best way. I and I do doubt if best method is well known at all. In one of the best music schools i read about had a good definition: If passer by can tell what song what was played by practising violin/cello/whatever player, then that is NOT practising. So 1 hours for person about 10-20 year of age, maybe 20 000 for 30's, not doable in 50's. At around 50 person start lose myelin at the rate he gets new. UNless you do try to learn something new. But yes learning is more tedious but there are no hard limits. Soft limits will get hard enough though. PP ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] 50k-100k patterns
2012/8/9 Mark Boon tesujisoftw...@gmail.com On Wed, Aug 8, 2012 at 5:10 AM, Jim O'Flaherty, Jr. jim_oflaherty...@yahoo.com wrote: So, if this great pruning has any real truth to it, it would likely be a large influencer in why both acquiring a language, an accent and learning intricate Go patterns prior to 6 are then hard coded at a very deep and abstract level into the resulting pruned network by the time the child has reached 10 years old. In effect, they have been imprinted deeply and subconsciously/intuitively in a way no adult can possibly match. This was my understanding as well. And it strokes with my observations. It is possible that this is the change in level of abstraction that the article is referring to when moving from 'amateur' to 'professional' level. But in that case it's only coincidence that it happened right at the age where this change happens and has nothing to do with the move to professional level. But the majority of people become 'professional' in pretty much anything at a later age. Only a small class of child prodigies reach professional level in their early teens. Well most people really never reach professional level on anything. Most SW developers I have met reach level adequate, but few are willing to invest the pain needed to become really good at it. As to language, I have never met a foreigner who learnt Dutch as an adult who didn't have an accent. Although Dutch sounds gutteral to the casual observer, it's in fact the vowels that seem to be impossible to get right for foreigners. Only when learnt as a child do they acquire Mark This can be that people are not really trying. I would leard Dutch right now I would probably have a bad accent as way forming vowels in Finnish and Dutch is vastly different. But let us assume that I would be willing invest about 10 000 hours into learning accent free Dutch I could get lot better results. And by this 10 000 hours it woudl mean dedicated goal oriented good quality training. But then againb who is. By the way I know one foreigner without accent speaking Finnish who learnt it at adult age, I guess that wont make rule that it is easy :) This last condition explains why most people do get to reasonable level at their profession but never excel. You can program 20 000 hours but unless you have clear goal of get better - not merely hoping that just doing what ever you are doing will get you there - one will not reach expert level Petri ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] scalability study
2011/6/17 Jean-loup Gailly jl...@gailly.net I have done precisely this. The reports of scalability death are greatly exaggerated, as you can see from the attached graph. To avoid self play benchmarks which are misleading, I tested Pachi against Fuego 1.1. Fuego Jean-loup Well this gives a biased solution. Wrong sample so to speak. Fuego will not create complex semeais and har read ishi-no-shita nakade shapes i.e opponent that puts no pressure to known problems . So you prove that agains opponent who does not play like human you do scale. But you advance the ladder of human players these small issues tend pop-up more often. Scaling measurement against strong humans is obviously bit hard. Just about only thing is letting different CPU machines play in KGS. Yes I do believe that pachi/Fuego will play better given more time. But It would scale better if there were better algorithm in place and part of that extra CPU would be used there. Just that exactly what to for it is bit murky. So I don't think that we get to 6 Dan EGF (8-9 Dan KGS?) with current programs just adding memory and CPU. Petri ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] [Computer-go ] Congratulations to Zen!
2011/6/16 Don Dailey dailey@gmail.com That's simple (at least I think). The scaling law converges at some point where the speed (or more thinking time) benefits little. It converges at perfect play. Thinking that it just happens to converge at the exact limits of Not by current playout algorithms. Sampling in a wrong way gives wrong statistics regardless how much you sample. Or so it was said in my Statistics and Probability course back in eighties. But I think it is still true. So 10 years from now 100 times faster computers will not create very good players by themselves. Part of that CPU must allocated to smarter algorithms. As I remember your previous post I would say this happened in chess as well? At least there are very good chess programs that do not do millions of evaluations/second http://en.wikipedia.org/wiki/Human-computer_chess_matches#Pocket_Fritz_4_.282009.29 So that is not scaling, thats programming I would say. Yes, partly scaling. Mobile phones in 2009 had pretty fast processors. Still nothing to compare desktops Don currentWhat is 9 x 15?And what is 20 minutes + 5 x 30, I don't understand the time controls.How much time is the human getting and how much time is the computer getting? I assume the computer is getting 13 and 28 right? So the extra time for the computer is just over 2X which isn't 9x15 is 15 seconds/move with max 9 overruns. So in the beginning you have 9 byo-yomi times left and if you spend 28 seconds thinking for the move you lose one. If think for instance 31 seconds you would have lost two periods. Basic Byo-yomi overtime. and second one is 20 minutes basic time and after that 5 periods for 30 seconds. http://en.wikipedia.org/wiki/Byoyomi Petri ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] May KGS bot tournament: 19x19, fast
2011/5/9 Nick Wedd n...@maproom.co.uk You are proposing that the tournament should start by pairing strong players with weak players, and claiming that this is more likely to result in the strongest player winning the tournament. I don't see it. Maybe itis easier if you think Swiss system as a cup. Which it for winners (assuming number of rounds being like 5 rounds for 32 players). If you pair strong players with strong ones, on second round you end having players still contending of winning that are both weak and strong. And some strong players dropped to competing for third place effectively. Swiss system is a cup with kinda consolation being played by the losers. And you have reliable a priori information then swiss is not not the best choice. MacMahon gets better results. Not fair for quickly advancinf players but nothing is perfect. You could simulate this easily by assuming that win probabilities follow exactly ELO estimates for example. Petri ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] February KGS bot tournament: 19x19
2011/1/30 Daniel Shawul dsha...@gmail.com Hello,My other problem is with implementation of 'byo-yomi time' . I am not familiar with this time control before. From my understanding there is an initial main time set where an engine can make as many moves at it wants. So does that mean if I make moves faster at that stage I get extra time or not ? What I implemented currently is just make (boardisze / 4) i.e half you your total moves in that time control and the remaining in byo-yomi mode. Doe that sound reasonable ? In Go related question Sensei's library is the 1st place to look for an answer. http://senseis.xmp.net/?ByoYomi Is not very usefull but it doers contain a link that is http://www.britgo.org/bgj/10643.html br, Petri ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Could a 'doubling dice'** encourage early resignation by programs?
2011/1/28 Álvaro Begué alvaro.be...@gmail.com On Thu, Jan 27, 2011 at 4:12 PM, Don Dailey dailey@gmail.com wrote: I don't understand your objection, Don. The side that is winning will at some point determine that the probability of winning the game is large enough (say, more than 80%) and it will propose doubling. At that point the losing side can resign and lose 1 point; resigning later (after accepting the doubling) costs 2 points. Álvaro. ___ Well I have an objection. Example you gave person offering the double already made a mistake. If one starts working with the cube, then it is needs to be done well. Optimal doubling point is when chance of winning is exactly 75%, then for the opponent it is indifferent whether he accept of rejects. And erring in direction doubling too late is far more serious than doubling too eraly. Also erring to direction of resigning instead of playing is usually far bigger mistake. So if a winds of war will make situation clear after next encounter, then the correct moment to double just before it. 55% chance of winning maybe enough in some cases,depends of how likely is the re-double. Term used used for this is volatility. Bigger the volatility smaller is the edge ususally required to double. And if there is no re-double 50%+epsilon is good enough to double and optimal is still as close to 75% you can get. So why we would introduce new set of skills to bve programmed into a game where it makes no sense? And when done optimally it will not shorten the games too often as correctly offered double should be accepted. Petri ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Semeais
2011/1/14 Robert Jasiek jas...@snafu.de Suggestion for MC: 1) Identify semeais by expert knowledge. 2) Divide the board into each semeai / non-semeai region. 3) Solve / represent each semeai by expert knowledge. 4) Do global MC with the change of special treatment for each semeai region. E.g., know the best black / white move in each semeai region and treat that region like just one intersection for the respective player to move (alternatively: as two intersections for either player). For other problems like life and death, endgame or game's first move, a similar Local Expert Knowledge approach could be used. There are complications of course: - dynamic region update - multiple threats on several regions - ko -- robert jasiek This would not help with problem of Semeais appearign during the MC simulation and expert system cannot be used during simulation as it cost too much. Petri ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Running automated bot matches?
http://www.lysator.liu.se/~gunnar/gtp/ id your source for GTP related stuff. Contains both Perl and Python script for running loads of games. I do wonder how you plan to make it fast though :) Petri 2011/1/11 Joona Kiiski joona.kii...@gmail.com Hi everyone, During the last week I've been examining sources of different open source go-engines (fuego, pachi, orego). Now I'd like to start making some simple modifications to some of them (not yet decided which one) and see how it goes (likely my 50 first tries will fail miserably, but it's okay). In computer chess programming, it's nowadays a widely accepted fact that only reasonable way to test changes is to run a huge number of test games between original and modified version. I assume that same applies also for go-programming. So, let us have open-source program X and slightly modified version it X'. What is the easiest way to run say 1000 super-fast games between them? I hope there already exists some scripts or programs to do this. My OS is Linux if it matters. Thanks for your help! ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] ManyFaces swindled of victory?
All my wind against ManyFaces tend to be swindles. It clearly better at certain things than I am, but when big groups get cut off and eyeless I can win fights where I should lose. I could easily find you a game where I won by winning a huge semeai where MFOG could have just by playing it out or by making eye/eyes earlier. So you don't need even many confusing fights. One big one will do. Well and in my cases a bit of luck as well. Petri 2010/11/13 terry mcintyre terrymcint...@yahoo.com This is a curious game - a 3 stone game betwixt ManyFaces and a 2 kyu player. It looks to my eye as if MFGO was well ahead in the middle game, yet managed to snatch defeat from the jaws of victory. Black has many unsettled groups; usually a stronger player causes much grief in situations like this, but black managed to turn around and swallow up not one but several white groups. Most interesting. I think this confirms an earlier thread - when there are several semeai on the board, MCT-based strategies get into difficulties. I think strong human players tend to have a better grasp of this particular fight is settled, no need to study it further until a liberty is removed (in which case respond now!); let's focus on this other fight instead . . . ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] new predictions?
In bachgammon being over champion may not be possible in measurable. Some bogts are better than other but amoumt games needed to measure it is already too much. To measure difference between top bot and top humans would need somthing like 500 point match and do doubt that anyone is willing go for that. So program mightget better but gains will too small to matter in go I would think is is reasonable estimat Petri - Alkuperäinen viesti - In 2002 H. J. van den Herik, J. W. H. M. Uiterwijk, and J. van Rijswijck. in Games solved: Now and in the future. Artificial Intelligence, 134:277–311 predicted that in 2010 Computers would have: Solved the games Awari Othello Checkers 8 x 8 Be over champion at Chess Draughts (10 x10) Scrabble Backgammon Lines of Action World champion at Go (9*9) Chinese chess Hex Amazons Grand master at Bridge Shogi Amateur at Go (19*19) I know Checkers (8x8) is solved but in how far were the other predictions correct? Also are new predictions made for 2020? ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] human complexity measure of games
No idea was to create amout of skill levels ina game. Level 0 would be total beginner level 1 would a player that can beat level 0 ålayer with 75 per cent probability with thinking times equalling chess tournament match 4 hours about then you could estimate skill levels from say elo max/min if my memory serves go has about 40 levels and chess 16. Checkes was about 8. Petri - Alkuperäinen viesti - Is this similar to measuring how many draws for games like chess and checkers? In checkers at the top levels, most games are draws. In chess, the top programs draw a lot more than they used to. I ran a match against my program Komodo and Stockfish and almost half of the games were draws. Unfortunately the meaning and difficulty of a draw varies from game to game and in some games a draw is not possible, in others a draw is much less likely because of the nature of the game. Don On Tue, Oct 26, 2010 at 2:52 AM, Olivier Teytaud olivier.teyt...@lri.frwrote: Dear all, I've been told recently that there are some works measuring how deep a game is as follows: - consider a fixed 0.5 p 1; - consider how many categories of people you can find such that the category number n wins with probability p against the catégory number n-1. Clearly, this is not so well defined - but it's interesting (at least to me :-) ). I've discussed with several people, some of them saying oh yes I remember I've already seen this, but nobody could remember the reference. Any precise reference or key word I could google ? Best regards, Olivier ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Pebbles opening learning
I presume this is for small boards? Hard to see in 19x19 that doing such statistics would be meaningful. Petri 2010/10/14 Brian Sheppard sheppar...@aol.com The Mogo team (among others) published extensive descriptions of how to write an opening library that learns. I recommend following their idea, rather than mine. :-) But if you don't have the time to do things properly, and you don't want to lose the exact same game over and over, then you might use Pebbles simple adaptive technique. Pebbles repeats moves that won the last time they were played. Specifically: 1) after every game, mark all positions in which the winner moved as winning-last-time. 2) In every game look up all successors of the current position. 3) Choose randomly among successors that were winning-last-time. 4) If no winning-last-time then search. This simple system is surprisingly robust. In particular, a) your program learns from opponents that defeat it. b) your program repeats moves that previously won. c) The selected moves have a strength that is lower-bounded by your program's native search strength, because i) A move either came from your own search, or ii) from a previous game in which an opponent defeated you. The weakness is in step 4: search. It is possible to have a winning position, but your search can't find the play. It is possible to upgrade step 4 so that the process is asymptotically optimal. That is, the system converges on maximal play. Pebbles hasn't actually implemented that feature, but I will have to implement that eventually in order to climb the ladder. Brian ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] Exhibition match
2010/10/4 Don Dailey dailey@gmail.com 1. boardsize could be a factor. 2. it may be more feasible in handicap games. 3. it may work better against humans for psychological reasons. 4. Maybe I did not stumble on the right approach when I did my experiments. I would not use dynamic komi in anything but handicap games. When computer playes white without some dynamic komi it will be on your face after just few moves. Which will usually/often fail. It is common wisdom that playing white in handicap games takes patience to win at the end. And as black it will play slack opening as it has too good chance to win . If you play against computer you will these two phenomens. I have and i clearly visible that they do start huge fights in handicap games where they really need excel to win. Petri ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
Re: [Computer-go] anti-pondering
Cut whenever you can. Leave cutting points behind. They will cut. All my won games against ManyFaces of Go are going about like this. In early game same unreasonable groupf MFOG dies. But inprocess I need leave cutting point behing or my groups are split in two. Then in Late middle game/eraly end game it will try to make them live. Usually in some fancy semeai. Which I sometimes win by two liberties sometime Ilose. But getting into semeai seems to easy enough :) And just for the record on my two cpu laptop all my winnings are on two stone handicap. Without the handicap I do get the semeais alright Petri 2010/9/15 Jeff Nowakowski j...@dilacero.org On 09/14/2010 04:36 PM, terry mcintyre wrote: From my observations of human-versus-bot games, a winning strategy against bots seems to be: Create several capturing races, even if you lose all of them. Is there an established, reliable way to create capturing races against bots? ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go ___ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go