Re: [computer-go] Congratulations to GNU and to MoGoBot19!
steve uurtamo wrote: > true, and a good point. time management other than attempting > to equally divide remaining time among the expected number of > remaining moves (which itself isn't so easy to estimate) is > complicated. But that is so much better than human time management! If the expected number of moves is based on applicable experience (including, maybe other games with the same opponent) and is updated as the move number increases, just the same as a 70 year old person has a longer life expectation than a 10 y.o. just for having survived 70 years, it will not experience serious problems. Humans, like myself, who do not take part in tournaments want to have at least 1/3 of our time unused to avoid "time pressure". Competitors may feel confident with just 5% of their time remaining, but that forces errors that would not have been played otherwise. Humans spend time looking at a clock, and are distracted by doing so. If the remaining time is small, they "reschedule" looking at the clock again soon, which adds extra pressure. Computers feel comfortable with any time settings, and no matter how naif the scheduling algorithm is, it will always be far better than human scheduling. Computers can safely approach using 99.999% of their time (asymptotically) and there is no other reason why a computer should lose on time than net lag. The reason for extra time (of any kind) is that humans are lost when they run out of time. Therefore, it clearly favors humans, because they would have lost that game. Jacques. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to GNU and to MoGoBot19!
On Tue, Jun 19, 2007 at 04:18:20PM -0700, steve uurtamo wrote: > true, and a good point. time management other than attempting > to equally divide remaining time among the expected number of > remaining moves (which itself isn't so easy to estimate) is > complicated. Even that has the complication of estimating the expected length of the game. Much easier to use something like 2% of the remaining time on each move. -Heikki -- 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] Congratulations to GNU and to MoGoBot19!
On Wed, Jun 20, 2007 at 11:55:47AM +0100, Jacques Basaldúa wrote: > Computers feel comfortable with any time settings, and no matter how > naif the scheduling algorithm is, it will always be far better than > human scheduling. Computers can safely approach using 99.999% of their > time (asymptotically) and there is no other reason why a computer should > lose on time than net lag. Depends on the program. Programs based on MC, UCT, or other such things can use as much or little time as they please. Programs like Gnu Go can only adjust some overall parameters, and then it does its thinking. The best it can do, is to adjust those again for the next move, if this one turned out to take more or less time. -Heikki -- 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] Congratulations to GNU and to MoGoBot19!
On Tue, 2007-06-19 at 17:54 -0700, steve uurtamo wrote: > actually, it's least complicated with sudden death. I don't agree. We are talking about time management from the humans point of view and the human player doesn't need to think as hard about playing quickly in order to save time for later. If he doesn't it's all about the clock and how he allocates his time. With Fischer time he just plays and the clock is more forgiving. You have to manage your time with any time control system but Fischer is the least distracting for the human. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to GNU and to MoGoBot19!
On Tue, 2007-06-19 at 18:16 -0700, steve uurtamo wrote: > sorry, i should have said that i think that it's least complicated > with sudden death. unless you mean to treat it internally as > if it's sudden death, but to use fisher time to make up for lag/delay. I'm talking about from the humans point of view. For a computer it's not quite as simple as sudden death. In sudden death you can always take a percentage of the remaining time. With Fischer I would probably just take the increment plus a percentage of the remaining time or perhaps just a percentage of the remaining time (but a fairly heavy percentage.) - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to GNU and to MoGoBot19!
On Wed, 2007-06-20 at 07:56 +0200, nando wrote: > Not sure this was mentioned before, but there's an interesting study > work presented at > http://senseis.xmp.net/?TimingSystemsRedux I just looked, after a quick scan it looks pretty good. It seems logical and there is no undo deference to tradition just for sentimental reasons but a pragmatic search/discussion for a practical time-control for go. After having though about this some more I think I would personally favor Fischer but with really small increments. Just enough of an increment to play obvious moves comfortably without rushing. On a computer server this would be 3-5 seconds. It should be whatever allows you to play an instant obvious move in a relaxed way and still have a second or two left over. Of course you should take this with a grain of salt since I'm not a go player. The idea is to have predictable round schedules similar to sudden death but not lose games due to made scrambles in simple endgames. There would still be time accumulating in cases where the moves really are trivial. Even for long tournaments it would be reasonable to make the increment no more than 5-10 seconds. (Of course the main time would be correspondingly longer.) This opinion is based on recent posts that claim once you are in byo-yomi the game is over anyway. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
[computer-go] CGOS issues
There is some server issue with CGOS and it has been going down unexpectedly from time to time. We suspect it's not due to the CGOS software but some kind of recent upgrade to the system. If CGOS goes down today, it may be down for a while since I will be out most of the time today. Just a heads up. But it did stay up all night, so perhaps the problem is solved. I'll probably put a restart mechanism in later today - at least to cover reboots of the server or total crashes of it. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to GNU and to MoGoBot19!
3-5 or 5-10 seconds is not a "relaxed" or "comfortable" pace for most human players. Byo yomi is usually set at 30 seconds per move. Canadian time controls might be "20 moves in five minutes", which is 300/20=15 seconds per -- players seem to find that they often are pushed to play the last five moves in ten seconds under Canadian rules. There's a lot of angst expressed by many players over time controls, but my thinking is "get over it" -- even though I myself have sometimes been burnt, it's just a cost associated with a) getting a tournament organized in a reasonable time frame, and b) not making your opponents cool their heels for long, unpredictable periods. Computers have longer attention spans than humans. In byo-yomi, the game isn't quite over, but it is predictable. One can usually determine the correct play comfortably inside of 30 seconds, probably even ten, but it is important to make that decision correctly, as one may otherwise lose a point here, a point there, thereby losing the game "in yose" ( in the endgame ). Terry McIntyre <[EMAIL PROTECTED]> They mean to govern well; but they mean to govern. They promise to be kind masters; but they mean to be masters. -- Daniel Webster - Original Message From: Don Dailey <[EMAIL PROTECTED]> To: computer-go Sent: Wednesday, June 20, 2007 5:16:23 AM Subject: Re: [computer-go] Congratulations to GNU and to MoGoBot19! On Wed, 2007-06-20 at 07:56 +0200, nando wrote: > Not sure this was mentioned before, but there's an interesting study > work presented at > http://senseis.xmp.net/?TimingSystemsRedux I just looked, after a quick scan it looks pretty good. It seems logical and there is no und[ue] deference to tradition just for sentimental reasons but a pragmatic search/discussion for a practical time-control for go. After having though about this some more I think I would personally favor Fischer but with really small increments. Just enough of an increment to play obvious moves comfortably without rushing. On a computer server this would be 3-5 seconds. It should be whatever allows you to play an instant obvious move in a relaxed way and still have a second or two left over. Of course you should take this with a grain of salt since I'm not a go player. The idea is to have predictable round schedules similar to sudden death but not lose games due to mad scrambles in simple endgames. There would still be time accumulating in cases where the moves really are trivial. Even for long tournaments it would be reasonable to make the increment no more than 5-10 seconds. (Of course the main time would be correspondingly longer.) This opinion is based on recent posts that claim once you are in byo-yomi the game is over anyway. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Take the Internet to Go: Yahoo!Go puts the Internet in your pocket: mail, news, photos & more. http://mobile.yahoo.com/go?refer=1GNXIC___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to GNU and to MoGoBot19!
On Wed, 2007-06-20 at 06:44 -0700, terry mcintyre wrote: > 3-5 or 5-10 seconds is not a "relaxed" or "comfortable" pace for most > human players. Byo yomi > is usually set at 30 seconds per move. Canadian time controls might be > "20 moves in five minutes", . Who said the pace is 5-10 seconds? There is a liberal main time-control that makes the pace of the game much more relaxed and comfortable. This post addresses the logic behind why Fischer time with small increments is best for both humans and computers. I stated my opinion on this before, but I'm giving more substance in this message. Read on ... The only complaint I've heard so far against sudden death is that players are losing in dead won positions because they cannot physically negotiate the interface to make moves. The small increment is designed to address that situation. But there are other good reasons for it too. If you try to address every possible situation, you basically have to keep adding time somewhere and do it in a way that takes control away from the player by giving it to the time-control mechanism in a fairly rigid, unforgiving way. Fischer is the only way I am aware of that gives the players as much control over their own time-allocation as possible without trying to manage time FOR the players too much, and yet recognizes that games can vary significantly in total length. byo-yomi and other systems such as the Bronstein clock assumes the player cannot manage his time and tries to manage it for him. At least to one degree or another. Fischer is just a special case of the time control used in other games where you are given so much time for N moves, and then you are given additional time for another batch of moves (but accumulated time is not taken away from you.) Here is a typical time control NOT using Fischer time: 40 moves in 30 minutes followed by ... 20 moves in 10 minutes followed by ... 20 moves in 10 minutes ... etc Fischer is exactly the same system but the number of moves is either 0 or 1. 0 moves in 15 minutes 1 move in 10 seconds 1 move in 10 seconds ... etc. What makes either system so good is that you are never penalized for moving quickly and it's understood that if a game lasts longer you need more time. Fischer just does this with finer granularity and makes the bookkeeping easier. The right parameters for Fischer time is whatever allows the highest quality of games in the shortest actual game time and of course these values can only be estimated or guessed at.I have estimated (perhaps incorrectly but based on many comments from the group and for other reasons too) that the highest quality games will result from giving a player a large pool of time for HIM to manage rather than trying to manage it for him with a small pool and large increment. Some of you may have noticed that the strongest CGOS programs spend most of their time on just the first few moves. Any time control that prevents this by playing big-brother to the time-allocating decisions will weaken these programs. This even includes Fischer time with a large increment. The larger the increment, the more control the system is imposing on you. All of these considerations together would seem to indicate that it is best to let the human have as much control as possible by allocating a large pool of initial time and keeping the increment pretty small (just what is needed to comfortably play the ending and perhaps a small bit more.) I believe this will lead to the strongest possible play on average for a given amount of time per round of play. Although the relative difference in computers vs humans vary with the time control mechanism used, what applies to one still applies to the other. For instance sudden death weakens a computer program too (just not relative to a human) because a computer doesn't know whether a game will last a long time or not and can't predict when the game starts getting easy. So it's all about who controls your time - YOU, or the time-control system. Byo-yomi is particularly heavy-handed in this regard. It has a very artificial mechanism for controlling the pace of the endings. It's like the mafia in that it makes you an offer you shouldn't refuse - make your moves in 30 seconds or less, or else Of course it will forgive you up to a point but it's not true forgiveness, it forgives but it doesn't forget. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to GNU and to MoGoBot19!
On Wed, 2007-06-20 at 11:12 -0400, Don Dailey wrote: > All of these considerations together would seem to indicate that it is > best to let the human have as much control as possible by allocating a > large pool of initial time and keeping the increment pretty small > (just > what is needed to comfortably play the ending and perhaps a small bit > more.) I believe this will lead to the strongest possible play on > average for a given amount of time per round of play. I would like to also mention that CGOS is an extreme example of Fischer clock time control with tiny increment. 1/4 second is silently added to every move on CGOS and the purpose was exactly the same as what we are talking about - to prevent time-loss when a game lasts too long and a program cannot "physically negotiate" the clock in the required time. Of course in the case of CGOS, this is essentially due to network lag where it is known that some strong programs have lost easily won games despite playing instantly on their local clocks. To be technically accurate, CGOS actually uses the Bronstein clock because time is not allowed to accumulate. Think of it as the first 1/4 second is for free, if you can move faster than that no time is charged against you but if it takes longer than 1/4 sec you get the 1/4 second as a time-lag compensation. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to GNU and to MoGoBot19!
> The right parameters for Fischer time is whatever allows the highest > quality of games in the shortest actual game time and of course these > values can only be estimated or guessed at.I have estimated (perhaps > incorrectly but based on many comments from the group and for other > reasons too) that the highest quality games will result from giving a > player a large pool of time for HIM to manage rather than trying to > manage it for him with a small pool and large increment. after reading the page linked to earlier (at sensei's), i was impressed to discover that the NYGC has had good success with fisher, experimenting with various controls. they switched from 10 minutes / 20 seconds fisher to 18 minutes / 15 seconds fisher. i'd be interested in trying a game or twelve this way at KGS, unless adding a time control would be a big pain. If it is (wms), shoot me the source and I'll be happy to add it. s. Shape Yahoo! in your own image. Join our Network Research Panel today! http://surveylink.yahoo.com/gmrs/yahoo_panel_invite.asp?a=7 ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Congratulations to GNU and to MoGoBot19!
On Wed, 2007-06-20 at 08:52 -0700, steve uurtamo wrote: > > The right parameters for Fischer time is whatever allows the highest > > quality of games in the shortest actual game time and of course these > > values can only be estimated or guessed at.I have estimated (perhaps > > incorrectly but based on many comments from the group and for other > > reasons too) that the highest quality games will result from giving a > > player a large pool of time for HIM to manage rather than trying to > > manage it for him with a small pool and large increment. > > after reading the page linked to earlier (at sensei's), i was impressed to > discover that the NYGC has had good success with fisher, experimenting > with various controls. they switched from 10 minutes / 20 seconds fisher > to 18 minutes / 15 seconds fisher. i'd be interested in trying a game or > twelve > this way at KGS, unless adding a time control would be a big pain. If it is > (wms), > shoot me the source and I'll be happy to add it. Yes, I saw that too. I suspect you could go even further and increase the main time more and the increment less. But I'm not a go player so my opinion on this may be less credible (of course I do not have bias from years of playing traditional time-controls either.) - Don > s. > > > > > > Shape Yahoo! in your own image. Join our Network Research Panel today! > http://surveylink.yahoo.com/gmrs/yahoo_panel_invite.asp?a=7 > > > ___ > 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/
