Re: [computer-go] Mega transposition table
On Fri, Jan 19, 2007 at 10:46:07AM -0500, Don Dailey wrote: On Fri, 2007-01-19 at 12:00 -0200, Mark Boon wrote: What ticked me off with the cuckoo method is apparently it can loop and a rehash is needed. Ouch, that better not happen very often! The whole point is that this cost is extremely tiny when amortized over the run time of the whole algorithm. In other words, the analysis of how efficient the algorithm considers this time. So the answer is that it is indeed a rare occurrence - a fraction of a percent of your run time. That also means that you have to keep enough data in each node to recalculate the hash keys from, for example the board position etc. Maybe it would be enough to keep the zobrist hash of the position, and use different bits of it for the hash key, or a different modulus, or something. -H -- Heikki Levanto In Murphy We Turst heikki (at) lsd (dot) dk ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] an idea for a new measure of a computer go program's rank.
On Sat, 2007-01-20 at 15:06 -0500, Don Dailey wrote: Years ago A player in the chess club kept beating me over the head with a non-standard opening move that was difficult to refute. I got sick of this, sat down in the privacy of my own home and didn't get back up until I discovered the correct response.In effect I consulted a much stronger player, myself, given a lot of extra time. I think I spent about 2 hours on this - so it was as if I consulted a player a few hundred ELO points stronger. I found a move I had no chance of finding in 20 or 30 seconds, even after repeated ad-hoc unstructured attempts. As soon as a started playing this move, my opponent stopped using it and he had to work harder to beat me. I forgot to mention an interesting addendum to this story.I was only about 1700 rated at the time and I later showed the position to a 2300 player - a good friend of mine. I had already figured out the correct response but out of curiosity I wanted to see how quickly the 2300 player would find the right response. I set up the position and he took a glance at it. He did a little analysis out loud and figured out the correct move but it took him about 30-60 seconds - it wasn't as quickly as I thought it would be. But it makes sense. He was some 600 ELO stronger than I was, so I would expect him to find it about 64 times faster if each doubling is worth 100 ELO. I don't know if you can apply the formula directly to a single move like this, but it was interesting nonetheless that it was roughly in the same ballpark. This very same master always analyzed his games and someone we analyzed them (and my games too) together. He often found moves he should have played that he didn't consider during the actual game. This was always after spending a great deal of time studying the position. You cannot tell me that thinking long and hard about a difficult move will not enable you to make a better one. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] an idea for a new measure of a computer go program's rank.
- Oorspronkelijk bericht - Van: Don Dailey [EMAIL PROTECTED] Datum: zaterdag, januari 20, 2007 9:06 pm Onderwerp: Re: [computer-go] an idea for a new measure of a computer go program's rank. Years ago A player in the chess club kept beating me over the head with a non-standard opening move that was difficult to refute. I got sick of this, sat down in the privacy of my own home and didn't get back up until I discovered the correct response.In effect I consulted a much stronger player, myself, given a lot of extra time. I think I spent about 2 hours on this - so it was as if I consulted a player a few hundred ELO points stronger. I found a move I had no chance of finding in 20 or 30 seconds, even after repeated ad-hoc unstructured attempts. As soon as a started playing this move, my opponent stopped using it and he had to work harder to beat me. It seems really odd to me that you are incapable of doing this in GO, or that the games are too different. If that's the case, then I prefer Chess, it is a far deeper game. I would find any game boring if it was so limited that there is nothing to think about that can't be seen in just a few moments. In my opinion in Go a game leaves the standard opening book very quickly, usually early in the opening. There are so many ways to play in the opening. If you opponent is trying to manipulate you into his favourite joseki(the taisha joseki for instance, with its proverbial 1000 variations), you have so many options to avoid it. But usually you just don't know what my opponent will play, so preparing for a particulal opponent is usually a waste of time. In my opinion, the difference is that in Go the possibity of variation is so great that a player is forced to rely on his own strength much earlier in the game than in Chess (in relation to the full length of a game). My level is 4d. For me the way to improve my results is studying professional games and Go problems. The aim is to get a very wide and general knowledge, more than a very deep knowledge of particular situations, because the level of variation in Go is so great. By improving you general knowledge of the game, you improve you ability to handle all those unique situations for which you cannot prepare in particular. Dave ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [spam probable] Re: [computer-go] Gnugo vs commercial programs
Hi Sylvain, On 1/10/07, Sylvain Gelly [EMAIL PROTECTED] wrote: So between the default level (8) and the level 16, there are 7% winning difference at around 50%, which is significant, but do not change by far the results Hiroshi posted. It is far less than 100 ELO right? I did not measure the thinking time of GnuGo level 16, but it seems quite long, and some games (at least 1, I don't remember) never finish after a lot of hours. Perhaps it is just a bug :). So I think using GnuGo level 8 is reliable (and for experiments much faster). If you have (or anyone else has) examples of .sgf-files with such extra-ordinary long thinking times for a single move, I would be interested in seeing them. (Send them to me, to gnugo-devel-at-gnu.org, or attach them at http://trac.gnugo.org/gnugo/ticket/160.) My suspicion is that most of them are related to explosion of branching factors in the local reading of ko fights - due to various reasons these are not very well controlled in GNU Go. Arend ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] an idea for a new measure of a computer go program's rank.
On Sat, 2007-01-20 at 21:55 +0100, [EMAIL PROTECTED] wrote: In my opinion in Go a game leaves the standard opening book very quickly, usually early in the opening. There are so many ways to play in the opening. If you opponent is trying to manipulate you into his favourite joseki(the taisha joseki for instance, with its proverbial 1000 variations), you have so many options to avoid it. But usually you just don't know what my opponent will play, so preparing for a particulal opponent is usually a waste of time. Yes, there are not volumes of exact memorized opening moves in Go and so you can't prepare against an opponent with specific memorized variations. Of course this has nothing to do with the point I was making about the relationship between thinking time and move quality. In my opinion, the difference is that in Go the possibity of variation is so great that a player is forced to rely on his own strength much earlier in the game than in Chess (in relation to the full length of a game). Yes, I would agree with this. Even Bobby Fischer noticed this and came up with a chess variant to render opening knowledge moot. My level is 4d. For me the way to improve my results is studying professional games and Go problems. The aim is to get a very wide and general knowledge, more than a very deep knowledge of particular situations, because the level of variation in Go is so great. By improving you general knowledge of the game, you improve you ability to handle all those unique situations for which you cannot prepare in particular. All interesting games require this - Go is not unique in this regard. I did not intend for anyone to think I was making a statement about the importance of memorizing openings or preparing for specific opponents. I get the feeling that you believed I was talking about this. I was responding to Ray Tayek who believes that he cannot produce higher quality moves no matter how much time he is given. That's not how it works for me. - Don Dave ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Gnugo vs commercial programs
Hello Arend, Unfortunately I don't have the log files anymore. I just remember that it (at least one) was a position with a lot of stones (not a starting position). Sylvain 2007/1/20, Arend Bayer [EMAIL PROTECTED]: Hi Sylvain, On 1/10/07, Sylvain Gelly [EMAIL PROTECTED] wrote: So between the default level (8) and the level 16, there are 7% winning difference at around 50%, which is significant, but do not change by far the results Hiroshi posted. It is far less than 100 ELO right? I did not measure the thinking time of GnuGo level 16, but it seems quite long, and some games (at least 1, I don't remember) never finish after a lot of hours. Perhaps it is just a bug :). So I think using GnuGo level 8 is reliable (and for experiments much faster). If you have (or anyone else has) examples of .sgf-files with such extra-ordinary long thinking times for a single move, I would be interested in seeing them. (Send them to me, to gnugo-devel-at-gnu.org, or attach them at http://trac.gnugo.org/gnugo/ticket/160.) My suspicion is that most of them are related to explosion of branching factors in the local reading of ko fights - due to various reasons these are not very well controlled in GNU Go. Arend ___ 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: [spam probable] Re: [computer-go] Gnugo vs commercial programs
I have such games. It was with a expermental version of Suzie, were Suzie played quite aggressive/over optimistic. Gnu-Go calculated very long, but won these games at the end completly When Suzie plays sound and wins or looses only be a small margin, Gnu-Go plays also with level 16 relative fast. I am currently in my private house in Austria, the games are on my computer in Germany (where I work currently during the week). I will send it on Monday. Chrilly - Original Message - From: Arend Bayer To: computer-go Sent: Saturday, January 20, 2007 11:09 PM Subject: Re: [spam probable] Re: [computer-go] Gnugo vs commercial programs Hi Sylvain, On 1/10/07, Sylvain Gelly [EMAIL PROTECTED] wrote: So between the default level (8) and the level 16, there are 7% winning difference at around 50%, which is significant, but do not change by far the results Hiroshi posted. It is far less than 100 ELO right? I did not measure the thinking time of GnuGo level 16, but it seems quite long, and some games (at least 1, I don't remember) never finish after a lot of hours. Perhaps it is just a bug :). So I think using GnuGo level 8 is reliable (and for experiments much faster). If you have (or anyone else has) examples of .sgf-files with such extra-ordinary long thinking times for a single move, I would be interested in seeing them. (Send them to me, to gnugo-devel-at-gnu.org, or attach them at http://trac.gnugo.org/gnugo/ticket/160.) My suspicion is that most of them are related to explosion of branching factors in the local reading of ko fights - due to various reasons these are not very well controlled in GNU Go. Arend -- ___ 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] an idea for a new measure of a computer go program's rank.
On Sat, 2007-01-20 at 15:34 -0700, Arend Bayer wrote: Hi Don, On 1/20/07, Don Dailey [EMAIL PROTECTED] wrote: If what you are saying is true, this is a waste of time. They should not be able to produce better quality moves than what they produce over the board. This has little to do with the question of whether you can improve a single move a lot by spending a lot of hours on it, but more with the fact that Go has many more reasonable moves in every opening position, so the game will leave your opening book preparation much quicker, especially compared to the overall length of the game. But I'm not talking about opening preparation. My point is all about just a few critical moves, not the majority of them. If you are given twice as much thinking time, there is bound to be 2 or 3 moves in a 300 move game where it makes a difference in the quality of those 2 or 3 moves. And that is worth 1 or more ranks of strength. snip: on improving a move in chess by spending many hours on it It seems really odd to me that you are incapable of doing this in GO, or that the games are too different. If that's the case, then I prefer Chess, it is a far deeper game. I would find any game boring if it was so limited that there is nothing to think about that can't be seen in just a few moments. I think of that in the opposite way. Go is such a deep game that in any position, there is a lot I will never be able to understand just by spending many hours on it. There are some things I may always misjudge that a professional will see immediately. If I think a group is weak and needs strengthening, but a pro just sees that it can never be attacked profitably, then that's not something where I can correct my mistaken thinking by spending many hours on the position. I believe this is all part of the strength/time relationship curve. If there is a huge disparity in playing strength, giving you a thousand times more thinking time won't be nearly enough to make up the gap. For instance ... Even when you double the speed of a chess playing computer, you add only a tiny amount of strength - so small it's not easily measured statistically. It's the same, I believe, with humans and probably why everyone here seems to believe what I'm saying is wrong, they think that I am implying that you can spend a few minutes on a move and play champion level. But if you are given twice as much thinking time, it's not going to turn you games from idiotic to brilliant. It will improve the (average) quality of your moves, but barely enough to notice. Having said that, I believe it's a lot more in GO based on some experiments I did with Steve Uurtamo in trying to get 19x19 CGOS ready. There is an ENORMOUS strength difference between programs that think twice as long - do 2X more monte carlo play-outs.Someone on this group (I can't remember who) correctly pointed out that a 19x19 has a lot of moves in it and so just a slight improvement in skill translates to a large winning percentage against even a slightly weaker opponent.This appears to be quite true. To put another perspective on it: If I had an hour for every move in a tournament game, I might play good EGF 5d level instead of average EGF 4d. That's a big difference from my perspective, but a small one when you compare it with the strength difference between me and a Korean who just became pro. This is understood. See what I said above.I don't really know how much 1 extra dan represents at this level - I think it translates to 200 or more ELO points. We can figure this out - what is the win expectancy of 5 dan over 4 dan without handicap? You said an hour per move - what are you comparing this against? 10 seconds per move? 1 minute per move? Arend ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] an idea for a new measure of a computer go program's rank.
Le dimanche 21 janvier 2007 01:23, Don Dailey a écrit : On Sat, 2007-01-20 at 15:34 -0700, Arend Bayer wrote: Hi Don, To put another perspective on it: If I had an hour for every move in a tournament game, I might play good EGF 5d level instead of average EGF 4d. That's a big difference from my perspective, but a small one when you compare it with the strength difference between me and a Korean who just became pro. This is understood. See what I said above.I don't really know how much 1 extra dan represents at this level - I think it translates to 200 or more ELO points. We can figure this out - what is the win expectancy of 5 dan over 4 dan without handicap? http://gemma.ujf.cas.cz/~cieply/GO/statev.html 4D 30.6% (out of 4000 games) Alain ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/