Re: [computer-go] Crazy Stone on 19x19 CGOS
The web site displays the wrong time-control. That will be confusing to people. Can you fix that? Unfortunately, I can not change things from where I am until wednesday. I'll fix all I can, starting from November 1st (or before if I can find a stable internet connection). Olivier ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] 19x19 CGOS
Hi all, For CGOS 19x19 I prefer a short time control (10min/game) because: 1) More quickly a more accurate rating can be established. I do think that the rating differences inbetween programs due to a shorter time setting do not change significantly (more than a few stones), while the rating difference of a newer program version (an update) within the pool can be shown with a better accuracy, due to the more games that will be played. 2) I am using my (single cpu) computer also for other things, and if I want to stop the cgos calculations I don't want to wait up to one hour before I can use it again. (It also takes longer before the first game starts). Just my opinion here. I noticed on cgos 19x19 that when crazystone stopped playing, it's name was not displayed on the cgos list anymore. What's is the cause of this? Edward. Date: Fri, 26 Oct 2007 07:32:42 +0900 From: [EMAIL PROTECTED] Subject: Re: [computer-go] 19x19 CGOS To: computer-go@computer-go.org I prefer shorter time control. The object I use cgos is to measure my program's performance against other programs. Cgos is not a tournament in any sense. It should be a tool for developers, I believe. Then, fairness is not so important because I can estimate my program's performace at longer time control easily. Most important thing for me is to know my program's rating _quickly_. I'd like to ask shorter time settings. -Hideki Olivier Teytaud: [EMAIL PROTECTED]: Ok for 30 minutes after the testing phase (for the tests I guess that 10 minutes is too long :-) ). For the moment I am trying to get the authorization of opening a port for socket connection - for the moment I guess only people in the same laboratory as me can connect to the machine, what is not a satisfactory behavior :-) Olivier ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ _ Nooit meer offline met Windows Live Messenger op je mobiele telefoon http://www.getlivemobile.nl/___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
I like the long time controls. I disagree about the rating difference, it makes a lot of difference because some programs respond to time more than others.It even makes a big difference in my own programs. 30 minutes is still way shorter than what is played in competitions. At the moment it will take especially long to establish a rating because almost every program is unrated. I doubt the current ratings are very accurate as a result of this.But you don't have to watch the games, just set it and forget it for a while. I am strongly considering an improvement where fast games are played to fill the time. I also think 30 is good because dual core and more will become more and more common. I have a dual core and it's wonderful - I can do something like play on cgos and also do other things with very little effect on the cgos game. - Don Edward de Grijs wrote: Hi all, For CGOS 19x19 I prefer a short time control (10min/game) because: 1) More quickly a more accurate rating can be established. I do think that the rating differences inbetween programs due to a shorter time setting do not change significantly (more than a few stones), while the rating difference of a newer program version (an update) within the pool can be shown with a better accuracy, due to the more games that will be played. 2) I am using my (single cpu) computer also for other things, and if I want to stop the cgos calculations I don't want to wait up to one hour before I can use it again. (It also takes longer before the first game starts). Just my opinion here. I noticed on cgos 19x19 that when crazystone stopped playing, it's name was not displayed on the cgos list anymore. What's is the cause of this? Edward. Date: Fri, 26 Oct 2007 07:32:42 +0900 From: [EMAIL PROTECTED] Subject: Re: [computer-go] 19x19 CGOS To: computer-go@computer-go.org I prefer shorter time control. The object I use cgos is to measure my program's performance against other programs. Cgos is not a tournament in any sense. It should be a tool for developers, I believe. Then, fairness is not so important because I can estimate my program's performace at longer time control easily. Most important thing for me is to know my program's rating _quickly_. I'd like to ask shorter time settings. -Hideki Olivier Teytaud: [EMAIL PROTECTED]: Ok for 30 minutes after the testing phase (for the tests I guess that 10 minutes is too long :-) ). For the moment I am trying to get the authorization of opening a port for socket connection - for the moment I guess only people in the same laboratory as me can connect to the machine, what is not a satisfactory behavior :-) Olivier ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Pas je zoekresultaten aan op JOUW wensen met Live.nl! Live.nl http://www.live.com/?mkt=nl-nl ___ 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] 19x19 CGOS
On Sun, 2007-10-28 at 11:54 +0100, Edward de Grijs wrote: Hi all, For CGOS 19x19 I prefer a short time control (10min/game) because: 1) More quickly a more accurate rating can be established. I agree with Don. 10 minutes sudden death is brutally short for 19x19. You are limiting the pool and strength of programs available for CGOS. If all you want is a quick and dirty rating for minor updates, why don't you just run your program against Gnu Go and/or MoGo at fast time settings on your own machine? Then when you think you have a stable and significant improvement, run your program on CGOS for a beefier test? This is how MoGo achieved dominance in 9x9. -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] 19x19 CGOS
Subject: RE: [computer-go] 19x19 CGOS From: [EMAIL PROTECTED] On Sun, 2007-10-28 at 11:54 +0100, Edward de Grijs wrote: Hi all, For CGOS 19x19 I prefer a short time control (10min/game) because: 1) More quickly a more accurate rating can be established. I agree with Don. 10 minutes sudden death is brutally short for 19x19. You are limiting the pool and strength of programs available for CGOS. Hi, maybe so, but can you name some programs which cannot cope with 10 minutes thinking time for 19x19? Note that my own program is a MC program which will play weaker in relation to for instance gnugo which the shorter thinking time, but I find that not important because as a programmer I want to see the relative progress over time. If all you want is a quick and dirty rating for minor updates, why don't you just run your program against Gnu Go and/or MoGo at fast time settings on your own machine? Then when you think you have a stable and significant improvement, run your program on CGOS for a beefier test? This is how MoGo achieved dominance in 9x9. This is just what I do with about 1 minute for each 13x13 game :-) In the past it happened that there were so many MoGo versions running on CGOS that it was questioned here in this mailing list if this could be reduced to create more diversity, if I remember this correctly. So CGOS was used by the MoGo team to get in impression about the rating of different updates in relation other programs then gnugo. And diversity is also the reason I like to test on CGOS. If it takes to long to establish a rating on CGOS I more often will use my own pool of programs (only gnugo for now) but then with different programs available, instead of establish a rating on CGOS which takes about 120 games or about 60 hours (estimation) of computer time. Maybe I am confused about the goals of CGOS? I thought that programmers could use it to get a good impression of improvements over time. And I also like to see the progress of other programs over time. I think this is also interesting to see for others. Edward. _ Live.nl: je eigen persoonlijk startpagina met nieuws en feeds die JIJ belangrijk vindt! http://www.live.com/getstarted.aspx___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] 19x19 CGOS
On Sun, 2007-10-28 at 15:59 +0100, Edward de Grijs wrote: Hi, maybe so, but can you name some programs which cannot cope with 10 minutes thinking time for 19x19? I'm working on my own program, and I don't want to be limited to 10 minutes for 19x19. I'll let others speak about their own programs. Maybe I am confused about the goals of CGOS? I thought that programmers could use it to get a good impression of improvements over time. Sure, to track improvements, but also to see which program is the strongest. Having the strongest program at a very fast speed is not as interesting as having the strongest program at a reasonable speed, for some definition of reasonable. I think getting a very fast rating on minor updates should not be the goal of CGOS -- you can do that on your own machine with Gnu Go, MoGo, and your own test suites. CGOS should be more like a continuous tournament to test major updates of programs. Waiting a day or two to get a rating at reasonable time controls then shouldn't be a big deal. That's my 2 cents, anyways. -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
Hi Edward, I can give you the goals of CGOS since I wrote CGOS for my own reasons. As a chess programmer I noticed that serious events and competitions were a huge impetus to making programming improvements. A lot of programmers told me the same thing, that despite the testing they did on the side, actual competitions seemed to reveal problems and bugs. So what I thought would be useful to the computer go community would be a forum for testing that could also stimulate competition and would have some meaning. In other words, I didn't make CGOS only as a way to test your program or even just to get a rating, but as a way to stimulate competition. That's a big key to most improvements in most fields, and nothing brings this out more than real competition with real numbers. I wanted it to mean something if your program makes it to the top 10 on CGOS, etc. You will probably notice that CGOS results have been used in papers written about computer GO, to verify that the techniques used in the paper have some validity. What I've always hated is unverifiable papers.There is a summary section near the end where the techniques being presented are experimentally verified with their own self-tests - which nobody else can usually verify because the program is not open to the public. CGOS is superb for that too - it's a public forum to expose your creations - good, bad or ugly, to the world. In computer chess, and I assume also in computer go, there is more status associated with games which are played at time controls us humans think are serious.Also, there is much more status associated with games that are public as opposed to private testing. Status is good in this context for computer go. It's why I made the choices I did and why I think longer time controls are better for the computer go community as a whole. I agree that there are reasonable arguments for faster time controls, I don't discount those reasons, but when all things are considered together, I think the reasons for having longer time controls make more sense. I believe even 30 minutes is fast, but it's a good compromise in my opinion. - Don Hi, maybe so, but can you name some programs which cannot cope with 10 minutes thinking time for 19x19? Note that my own program is a MC program which will play weaker in relation to for instance gnugo which the shorter thinking time, but I find that not important because as a programmer I want to see the relative progress over time. Maybe I am confused about the goals of CGOS? I thought that programmers could use it to get a good impression of improvements over time. And I also like to see the progress of other programs over time. I think this is also interesting to see for others. Edward. Publiceer JOUW leven online met Windows Live Spaces: weblog, foto, video en muziek! Het is gratis! Het is gratis! http://clk.atdmt.com/MSN/go/msnnksac003001msn/direct/01/?href=http://www.imagine-msn.com/spaces ___ 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] 19x19 CGOS
Would anyone be interested in a highly configurable version 11 with gtp interface? Version 11 has a set of parameters that control the searching that I can easily read from a file. /* LEVELS:1 2 3 4 5 6 7 8 9 10 */ int maxmoves[NUMLEVELS] = /* maximum number of moves to try on full board */ { 0, 2, 4, 5, 6, 8,10,12, 15, 20, 28 }; /* lots, so in endgame can look at lots of moves */ int maxvariations[NUMLEVELS] = /* max number of leafs per move tried */ { 0, 1, 1, 2, 2, 3, 4, 5, 7, 10, 13 }; char maxscorebrdepth[NUMLEVELS] = /* max depth for any branches in getscore scorebestmove */ { 0, 0, 0, 1, 1, 2, 2, 2, 2, 3, 3 }; char maxscoredepth[NUMLEVELS] = /* max depth for getscore */ { 0, 0, 0, 1, 1, 2, 2, 3, 3, 3, 4 }; int maxlifecalls[NUMLEVELS] = /* total evaluations, should be around maxmoves*maxvariations */ { 0, 5, 9,13,20,30,45,65,95,200,400 }; /* LEVELS:1 2 3 4 5 6 7 8 9 10 */ unsigned char taclibs[NUMLEVELS] = /* max liberties in a tactical fight */ { 0, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4 }; unsigned char eyetaclibs[NUMLEVELS] = /* max liberties for eye diagonal */ { 0, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3 }; int cancapsize[NUMLEVELS] = /* size of search in canbecaptured */ { 0, 7,10,15,20,30,40,60,80,110,150 }; unsigned char eyecapsize[NUMLEVELS] = /* size of search for eyes diags */ { 0, 2, 3, 4, 5,10, 15,20,25,30, 40 }; unsigned char eyecapdepth[NUMLEVELS] = /* depth of search for eyes diags */ { 0, 2, 2, 3, 3, 4, 5, 6, 6, 6, 6 }; unsigned char conncapsize[NUMLEVELS] = /* size of search for connections */ { 0, 4, 6, 8,10, 20,30,40,55,80,100 }; unsigned char conncapdepth[NUMLEVELS] = /* depth of search for connections */ { 0, 2, 3, 4, 5, 7, 8, 9,10, 12, 14 }; char mvmost[NUMLEVELS] =/* number of moves considered for ladder at each ply */ { 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3 }; char eyemost[NUMLEVELS] =/* number of moves considered for ladder at each ply */ { 0, 1, 1, 1, 2, 2, 2, 2, 2, 3, 3 }; char connmost[NUMLEVELS] =/* number of moves considered for ladder at each ply */ { 0, 1, 1, 1, 2, 2, 2, 2, 2, 3, 3 }; int maxbranchdepth[NUMLEVELS] = /* maximum depth for branches in tactical move tree (unless move values are close) */ { 0, 1, 2, 2, 3, 3, 4, 4, 4, 4, 4 }; int maxtacdiff[NUMLEVELS] = /* maximum difference between best tac move and this move*/ { 0,16,16,16,32,64,64,96,120,180,250 }; int mintacval[NUMLEVELS] = /* minimum value move has to be considered tacticaly */ { 0, 0, 0, 0, 0,-10,-10, -10,-16,-20,-31 }; int numpotmoves[NUMLEVELS] = /* Number of moves to read for adpot() to capture group */ { 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 2 }; /* LEVELS:1 2 3 4 5 6 7 8 9 10 */ int maxjosvariations[NUMLEVELS] = /* max number of joseki variations - endpoints per first level joseki move */ { 0, 1, 1, 1, 1, 1, 2, 3, 3, 4, 6 }; int maxpatvariations[NUMLEVELS] = /* max number of pattern variations per move */ { 0, 1, 1, 1, 2, 2, 2, 3, 4, 5, 6 }; int maxjosbranches[NUMLEVELS] = /* max number of joseki variations per move at depth 1 */ { 0, 1, 1, 1, 1, 1, 2, 2, 2, 2, 3 }; unsigned char mdist[NUMLEVELS] = /* distance to radiate influence from live groups */ { 0, 7, 8, 9, 10, 11, 11, 12, 12, 13, 13 }; /* Fights: no fight reading below level 5 */ /* LEVELS:1 2 3 4 5 6 7 8 9 10 */ int maxfightbranches[NUMLEVELS] = /* max number of fight variations per move */ { 0, 1, 1, 1, 1, 1, 2, 2, 2, 3, 3 }; char maxfightdepth[NUMLEVELS] = /* max depth for reading fight */ { 0, 0, 0, 1, 1, 2, 3, 4, 6, 7, 9 }; int maxfightbrdepth[NUMLEVELS] = /* max depth for branches in reading fight */ { 0, 0, 0, 1, 1, 1, 2, 3, 4, 5, 5 }; int maxsemdiff[NUMLEVELS] = /* maximum difference between best semeai move and this move*/ { 0, 8,16,24,32,40,50,60,80,90,100 }; i have a copy of 11. is there any way to crank it up other than level 10. maybe a config file somewhere? have you considered a highly configurable version 12 for some of us on the list? ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] 19x19 CGOS
At 09:27 AM 10/28/2007, you wrote: Would anyone be interested in a highly configurable version 11 with gtp interface? ... i'll buy one. thanks --- vice-chair http://ocjug.org/ ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
RE: [computer-go] 19x19 CGOS
While I don't own a copy of Many Faces (and probably won't for a while), what you suggest would be a big help to my use of it. On Sun, 2007-10-28 at 09:27 -0700, David Fotland wrote: Would anyone be interested in a highly configurable version 11 with gtp interface? Version 11 has a set of parameters that control the searching that I can easily read from a file. /* LEVELS:1 2 3 4 5 6 7 8 9 10 */ int maxmoves[NUMLEVELS] = /* maximum number of moves to try on full board */ { 0, 2, 4, 5, 6, 8,10,12, 15, 20, 28 }; /* lots, so in endgame can look at lots of moves */ int maxvariations[NUMLEVELS] = /* max number of leafs per move tried */ { 0, 1, 1, 2, 2, 3, 4, 5, 7, 10, 13 }; char maxscorebrdepth[NUMLEVELS] = /* max depth for any branches in getscore scorebestmove */ { 0, 0, 0, 1, 1, 2, 2, 2, 2, 3, 3 }; char maxscoredepth[NUMLEVELS] = /* max depth for getscore */ { 0, 0, 0, 1, 1, 2, 2, 3, 3, 3, 4 }; int maxlifecalls[NUMLEVELS] = /* total evaluations, should be around maxmoves*maxvariations */ { 0, 5, 9,13,20,30,45,65,95,200,400 }; /* LEVELS:1 2 3 4 5 6 7 8 9 10 */ unsigned char taclibs[NUMLEVELS] = /* max liberties in a tactical fight */ { 0, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4 }; unsigned char eyetaclibs[NUMLEVELS] = /* max liberties for eye diagonal */ { 0, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3 }; int cancapsize[NUMLEVELS] = /* size of search in canbecaptured */ { 0, 7,10,15,20,30,40,60,80,110,150 }; unsigned char eyecapsize[NUMLEVELS] = /* size of search for eyes diags */ { 0, 2, 3, 4, 5,10, 15,20,25,30, 40 }; unsigned char eyecapdepth[NUMLEVELS] = /* depth of search for eyes diags */ { 0, 2, 2, 3, 3, 4, 5, 6, 6, 6, 6 }; unsigned char conncapsize[NUMLEVELS] = /* size of search for connections */ { 0, 4, 6, 8,10, 20,30,40,55,80,100 }; unsigned char conncapdepth[NUMLEVELS] = /* depth of search for connections */ { 0, 2, 3, 4, 5, 7, 8, 9,10, 12, 14 }; char mvmost[NUMLEVELS] =/* number of moves considered for ladder at each ply */ { 0, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3 }; char eyemost[NUMLEVELS] =/* number of moves considered for ladder at each ply */ { 0, 1, 1, 1, 2, 2, 2, 2, 2, 3, 3 }; char connmost[NUMLEVELS] =/* number of moves considered for ladder at each ply */ { 0, 1, 1, 1, 2, 2, 2, 2, 2, 3, 3 }; int maxbranchdepth[NUMLEVELS] = /* maximum depth for branches in tactical move tree (unless move values are close) */ { 0, 1, 2, 2, 3, 3, 4, 4, 4, 4, 4 }; int maxtacdiff[NUMLEVELS] = /* maximum difference between best tac move and this move*/ { 0,16,16,16,32,64,64,96,120,180,250 }; int mintacval[NUMLEVELS] = /* minimum value move has to be considered tacticaly */ { 0, 0, 0, 0, 0,-10,-10, -10,-16,-20,-31 }; int numpotmoves[NUMLEVELS] = /* Number of moves to read for adpot() to capture group */ { 0, 0, 0, 0, 0, 0, 1, 1, 2, 2, 2 }; /* LEVELS:1 2 3 4 5 6 7 8 9 10 */ int maxjosvariations[NUMLEVELS] = /* max number of joseki variations - endpoints per first level joseki move */ { 0, 1, 1, 1, 1, 1, 2, 3, 3, 4, 6 }; int maxpatvariations[NUMLEVELS] = /* max number of pattern variations per move */ { 0, 1, 1, 1, 2, 2, 2, 3, 4, 5, 6 }; int maxjosbranches[NUMLEVELS] = /* max number of joseki variations per move at depth 1 */ { 0, 1, 1, 1, 1, 1, 2, 2, 2, 2, 3 }; unsigned char mdist[NUMLEVELS] = /* distance to radiate influence from live groups */ { 0, 7, 8, 9, 10, 11, 11, 12, 12, 13, 13 }; /* Fights: no fight reading below level 5 */ /* LEVELS:1 2 3 4 5 6 7 8 9 10 */ int maxfightbranches[NUMLEVELS] = /* max number of fight variations per move */ { 0, 1, 1, 1, 1, 1, 2, 2, 2, 3, 3 }; char maxfightdepth[NUMLEVELS] = /* max depth for reading fight */ { 0, 0, 0, 1, 1, 2, 3, 4, 6, 7, 9 }; int maxfightbrdepth[NUMLEVELS] = /* max depth for branches in reading fight */ { 0, 0, 0, 1, 1, 1, 2, 3, 4, 5, 5 }; int maxsemdiff[NUMLEVELS] = /* maximum difference between best semeai move and this move*/ { 0, 8,16,24,32,40,50,60,80,90,100 }; i have a copy of 11. is there any way to crank it up other than level 10. maybe a config file somewhere? have you considered a highly configurable version 12 for some of us on the list? ___ 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] 19x19 CGOS
Hi all, Jeff Nowakowski: [EMAIL PROTECTED]: On Sun, 2007-10-28 at 11:54 +0100, Edward de Grijs wrote: Hi all, For CGOS 19x19 I prefer a short time control (10min/game) because: 1) More quickly a more accurate rating can be established. I agree with Don. 10 minutes sudden death is brutally short for 19x19. You are limiting the pool and strength of programs available for CGOS. If all you want is a quick and dirty rating for minor updates, why don't you just run your program against Gnu Go and/or MoGo at fast time settings on your own machine? Then when you think you have a stable and significant improvement, run your program on CGOS for a beefier test? This is how MoGo achieved dominance in 9x9. We need thousands of games to get a few percent of standard deviation on both 9x9 and 19x19. So, of course I do what you wrote. When a game on cgos takes about one hour, a handred games take a handred hours, ie, four days. When I want to know my program's winning rate against a paticular program that I don't have in hand, it takes four days times the number of programs running on cgos at an average. So, it takes a few weeks in total which is tooo long for me. In constrast, I can guess my program's scalability by local competitions against GNU Go and/or MoGo. About fairness, as classical programs including GNU Go and ManyFaces need about ten minutes for their best performace, why do you give other (Monte Carlo) programs thirty minutes? I argue ten or fifteen minutes setting is enough and better for many developers than thirty minutes. -Hideki -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
About fairness, as classical programs including GNU Go and ManyFaces need about ten minutes for their best performace, why do you give other (Monte Carlo) programs thirty minutes? What time control do they use in serious tournaments?Do you consider them fair or unfair? - Don I argue ten or fifteen minutes setting is enough and better for many developers than thirty minutes. -Hideki -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ 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] 19x19 CGOS
I think I agree with Ed, but I also see and appreciate the arguments you give as well. I also like to watch CGOS games to evaluate my bot, but 1 hour per game is somewhat past my attention span (for real go games too). In all likelihood, I'll probably stick to 9x9 for most of my stuff (largest reason, games finish faster), and only switch to 19x19 when I'm good enough at the basics to be near the top of 9x9. If too many of us do that, 19x19 may suffer a similar fate to what it did in the past (that might not be true with Many Faces and others joining this time around). I think the idea of multiplexing in many 9x9 games between a few 19x19 games is a good feature that I'd likely take advantage of... Probably not enough to get my bot out of the yellow, but enough to get a flavor of how it performs on 19x19. On Sun, 2007-10-28 at 12:03 -0400, Don Dailey wrote: Hi Edward, I can give you the goals of CGOS since I wrote CGOS for my own reasons. As a chess programmer I noticed that serious events and competitions were a huge impetus to making programming improvements. A lot of programmers told me the same thing, that despite the testing they did on the side, actual competitions seemed to reveal problems and bugs. So what I thought would be useful to the computer go community would be a forum for testing that could also stimulate competition and would have some meaning. In other words, I didn't make CGOS only as a way to test your program or even just to get a rating, but as a way to stimulate competition. That's a big key to most improvements in most fields, and nothing brings this out more than real competition with real numbers. I wanted it to mean something if your program makes it to the top 10 on CGOS, etc. You will probably notice that CGOS results have been used in papers written about computer GO, to verify that the techniques used in the paper have some validity. What I've always hated is unverifiable papers.There is a summary section near the end where the techniques being presented are experimentally verified with their own self-tests - which nobody else can usually verify because the program is not open to the public. CGOS is superb for that too - it's a public forum to expose your creations - good, bad or ugly, to the world. In computer chess, and I assume also in computer go, there is more status associated with games which are played at time controls us humans think are serious.Also, there is much more status associated with games that are public as opposed to private testing. Status is good in this context for computer go. It's why I made the choices I did and why I think longer time controls are better for the computer go community as a whole. I agree that there are reasonable arguments for faster time controls, I don't discount those reasons, but when all things are considered together, I think the reasons for having longer time controls make more sense. I believe even 30 minutes is fast, but it's a good compromise in my opinion. - Don Hi, maybe so, but can you name some programs which cannot cope with 10 minutes thinking time for 19x19? Note that my own program is a MC program which will play weaker in relation to for instance gnugo which the shorter thinking time, but I find that not important because as a programmer I want to see the relative progress over time. Maybe I am confused about the goals of CGOS? I thought that programmers could use it to get a good impression of improvements over time. And I also like to see the progress of other programs over time. I think this is also interesting to see for others. Edward. Publiceer JOUW leven online met Windows Live Spaces: weblog, foto, video en muziek! Het is gratis! Het is gratis! http://clk.atdmt.com/MSN/go/msnnksac003001msn/direct/01/?href=http://www.imagine-msn.com/spaces ___ 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] 19x19 CGOS
I added a copy of Many Faces of Go running at level 1 (with almost no search) to add some variety for the weak programs. This version looks at the top 2 suggestions from the move generator, does a 1 ply search without quiescence, does a full board evaluation for each, and picks the best one. Late in the game it includes a pass move in the search, so it does 3 evaluations rather than 2. David -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Olivier Teytaud Sent: Friday, October 26, 2007 9:15 AM To: computer-go Subject: [computer-go] 19x19 CGOS The cgos 19x19 server is seemingly ok, the port 6919 is now opened for all the universe. The name of the machine is cgos.lri.fr (and not pc5-120.lri.fr as previously). The port is 6919. It is 19x19, 10 minutes per side for testing; I will move to something longer later (depending on what people prefer, I'll do a weighted average of durations suggested on the mailing list :-) ). http://www.lri.fr/~teytaud/cgosStandings.html Unfortunately, I'll be away from my email from tomorrow to wednesday and will not be able to correct the troubles that people will almost surely find in this installation; sorry for that. The installation is a bit complicated in order to avoid troubles due to the firewall and I am almost sure that some troubles will appear very soon :-) All comments welcome (in particular in the next hours as I am still close to my computer a few hours :-) ). [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] 19x19 CGOS
I'm working on MFGO 12 and I'd like 30 minutes so I can test against a variety of programs at tournament time limits. I don't need hundreds of games to tune, since my program is knowledge based. I'm not just changing parameters and seeing what happens. I'm looking for bad moves and adding knowledge. David About fairness, as classical programs including GNU Go and ManyFaces need about ten minutes for their best performace, why do you give other (Monte Carlo) programs thirty minutes? I argue ten or fifteen minutes setting is enough and better for many developers than thirty minutes. -Hideki ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
Don Dailey: [EMAIL PROTECTED]: About fairness, as classical programs including GNU Go and ManyFaces need about ten minutes for their best performace, why do you give other (Monte Carlo) programs thirty minutes? What time control do they use in serious tournaments?Do you consider them fair or unfair? Those settings are established earlier, ie, when we had poor pcs. As David mentioned, we had much less cpu power and needed 30 minutes for best performance. When we use almost the same method the absolute value of time setting is not a problem. But now we have two different approaches, classical and MC, too long time setting gives some advantage to MC programs. From the view point of innovations, however, it's not to be said unfair. When comparing performaces of several implementations of different approaches, ie, MC and classical, one scales better for time and the other is not, _at a moment_, it may be better to set the time being enough for classical programs. -Hideki - Don I argue ten or fifteen minutes setting is enough and better for many developers than thirty minutes. -Hideki -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ 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/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
Regarding Don Dailiey's rationale for CGOS and 30-minute (or longer) time controls: a hearty AMEN! The goal here is to improve the quality of play - not merely at blitz pace, but at slower rate more comparable to the pace of humans. Some older programs peak at 10 minutes for a 19x19 game; they were designed to run on 50 MHz machines, a decade back. It might be that, for the short term, variations of Monte Carlo on quad cpus can make better use of 30 minute or longer time controls than the traditional single-threaded programs. What better incentive to the developers to try multi-threading? They'll need a strong incentive to do so, since it is a non-trivial step. But consumers of Go programs will benefit from stronger, more interesting competition. Don's idea of packing in blitz games between the longer games makes a lot of sense; it would enable a second track for those who want results more quickly. Many thanks to Don and everyone else for making CGOS possible! 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 __ 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/
[computer-go] CGOS 19 TC
There is a lot of talk about time controls, and would like to add my input. I agree we should have longer time controls. I'm in the very early stages of my Go engine. With my current time line I dont anticipate having a running engine for at least a year. My design is a good bit different than the engines I've seen. Guess my question is, how resource intensive is the server code? Could we split it? Have the server running on one port for 10min blitz, and another port with 30+minute games? My only concern is fragmenting the amount of players. Not sure how many active people play, but if it's little it might make the pool to small. What are all of your thoughts? -Josh ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On Oct 28, 2007, at 11:16 AM, terry mcintyre wrote: Don's idea of packing in blitz games between the longer games makes a lot of sense; it would enable a second track for those who want results more quickly. I too like that idea. Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] CGOS 19 TC
My choice is 30 min per side, but I understand that some people are more interested in 10 min per side, so I suggest that the cgos protocol have a requested time option. If a player requests 10 min then there is an attempt to match it against another player requesting the same time, likewise for 30 min. If no time is specified then the match is with whichever player is available next and most appropriate for the pairings. It may make the pairing code a little more complicated, but it allows for choice. Cheers, David On 28, Oct 2007, at 11:26 AM, Joshua Shriver wrote: There is a lot of talk about time controls, and would like to add my input. I agree we should have longer time controls. I'm in the very early stages of my Go engine. With my current time line I dont anticipate having a running engine for at least a year. My design is a good bit different than the engines I've seen. Guess my question is, how resource intensive is the server code? Could we split it? Have the server running on one port for 10min blitz, and another port with 30+minute games? My only concern is fragmenting the amount of players. Not sure how many active people play, but if it's little it might make the pool to small. What are all of your thoughts? -Josh ___ 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] 19x19 CGOS
I think a lot of the early CGOS ratings were (are?) very skewed. It had two anchors at a (arbitrary) fixed distance of 600 but of almost the same strength (win-rate 49-51%). It will take several days to overcome that. Chrisotph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
Couldn't there just be two servers? There were multiple volunteers. A server with long games might draw more viewers but fewer participants. Shorter games would be more helpful for those of us working on weak 19x19 programs that other people are less interested in anyway. - Dave Hillis -Original Message- From: terry mcintyre [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Sun, 28 Oct 2007 2:16 pm Subject: Re: [computer-go] 19x19 CGOS Regarding Don Dailiey's rationale for CGOS and 30-minute (or longer) time controls: a hearty AMEN! The goal here is to improve the quality of play - not merely at blitz pace, but at slower? rate more comparable to the pace of humans. Some older programs? peak at 10 minutes for a 19x19? game; they were designed to run on? 50 MHz machines,? a decade back. It might be that, for the short term, variations of Monte Carlo on? quad? cpus? can make better use of? 30 minute or longer time controls than the traditional single-threaded programs. What better incentive to the developers to? try multi-threading?? They'll need a strong incentive to do so, since it is a non-trivial step. But consumers of Go programs will benefit from stronger, more interesting competition. Don's idea of packing in blitz games between the longer games makes a lot of sense; it would enable a second track for those who want results more quickly. Many thanks to Don and everyone else for making CGOS possible! 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 __ 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/ Check Out the new free AIM(R) Mail -- Unlimited 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] 19x19 CGOS
If I combine some reactions so far I understand that the main motivation to have 30min/game or longer time controls is that that is more comparable to the pace of humans, and that is is more easy for some new programs (not MC based) I can imagine that some humans will argue that blitz ratings are not to be trusted, but I think that they can be. For MC programs this only mean that they are about 1 or 2 stones weaker (at 10 min. instead of 30 min) in relation to (already fast) traditional programs. The resulting difference against humans I do not know. If someone builds a different engine, (not MC) I can image that the time can be important, but only if the program needs that kind of time control. For MC programs one have to realise that the difference between the best program and the average programs is about 8 stones or more, (a very rough estimation of mine). So the time control only accounts here for only about 2 stones, which will not help the new programs to perform much better. Those large differences could be corrected by introducing handicap stones, but I realise that will not be easy to combine with an elo rating scale. This brings me to the beginning of the newly started CGOS 19x19: I thought one of the first goals was to get an impression between the strenght of MFGO and CrazyStone. I do not see any discussions related to that, while it is very interesting what is happening (or has happened already) Crazystone was about 2000 elo, mfgo is about 1800. The CrazyStone row has dissapeared because not enough games were played, so there will be a larger standard deviation around those values (I expect a 1 sigma value of about 50 elo. It would be interesting to incluse those numbers on every row (Don?)) What I think is happening is that Crazystone seems to be about 7 stones (or more) stronger than MFGO at these time controls, when CrazyStone will use 6 cores or so (which Remi has used in the past). How: Crazystone used about 3 minutes for each game, one using 1 cpu, so it has handicapped himself (maybe Remi is nice to David?). When using almost the full 30 minutes it will be about 3 stones or 300 elo stronger. Combined with 6 cores that will be another 2 or 3 stones or 250 elo. So we are looking at Crazystone 2550 elo, MFGO 1800 elo which roughly corresponds to 7 stones! Just food for though, and my opinion and my rough estimates which will be erratic (with a certain deviation:-) Edward _ De mooiste afbeeldingen van Angelina Jolie vind je met Live Search http://search.live.com/images/results.aspx?q=angelina%20jolieFORM=QBIR___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS (CS vs MFG)
Edward de Grijs wrote: The CrazyStone row has dissapeared because not enough games were played, so there will be a larger standard deviation around those values (I expect a 1 sigma value of about 50 elo. It would be interesting to incluse those numbers on every row (Don?)) Uncertainty about the rating is much more. Also I stopped CS-8-26-10k-1CPU, because it was deterministic, and so is Many Faces. So they were playing the same game again and again. I have now connected a parallel version, running on two cores, which makes it random. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
I'm always going to tend to favor longer time controls. I don't think anyone here can reasonably argue that the quality of the games goes up with faster time controls - it's just the opposite.And given a choice between lower and higher quality games, I would tend to favor higher quality games. If longer time controls actually favor a certain type of program, then we have a choice: 1. Choose a time control that favors programs that excel at time controls that produce weaker play. 2. Choose a time control that favors programs that excel at time controls that produce stronger play. You obviously can't choose a time control that works best for any kind of program, but you certainly don't want to favor programs that can only win if they play quickly, if our goal is to encourage the development of the strongest possible programs. - Don Hideki Kato wrote: Don Dailey: [EMAIL PROTECTED]: About fairness, as classical programs including GNU Go and ManyFaces need about ten minutes for their best performace, why do you give other (Monte Carlo) programs thirty minutes? What time control do they use in serious tournaments?Do you consider them fair or unfair? Those settings are established earlier, ie, when we had poor pcs. As David mentioned, we had much less cpu power and needed 30 minutes for best performance. When we use almost the same method the absolute value of time setting is not a problem. But now we have two different approaches, classical and MC, too long time setting gives some advantage to MC programs. From the view point of innovations, however, it's not to be said unfair. When comparing performaces of several implementations of different approaches, ie, MC and classical, one scales better for time and the other is not, _at a moment_, it may be better to set the time being enough for classical programs. -Hideki - Don I argue ten or fifteen minutes setting is enough and better for many developers than thirty minutes. -Hideki -Jeff ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ -- [EMAIL PROTECTED] (Kato) ___ 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/ -- [EMAIL PROTECTED] (Kato) ___ 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] CGOS 19 TC
My thoughts are that it would fragment the players, we would get much less activity on either server and we need lot's of variety. However, I'm really liking the idea of playing quick 19x19 matches between rounds. It's the best of both worlds.I would arrange it so that this would never prevent you from playing the long games too. And you could choose to only play in the fast or slow games or both. - Don Joshua Shriver wrote: There is a lot of talk about time controls, and would like to add my input. I agree we should have longer time controls. I'm in the very early stages of my Go engine. With my current time line I dont anticipate having a running engine for at least a year. My design is a good bit different than the engines I've seen. Guess my question is, how resource intensive is the server code? Could we split it? Have the server running on one port for 10min blitz, and another port with 30+minute games? My only concern is fragmenting the amount of players. Not sure how many active people play, but if it's little it might make the pool to small. What are all of your thoughts? -Josh ___ 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] 19x19 CGOS
Hi Dave, Two servers is easy, but 1 server is better.The plan is that I will combine fast and slow games into one server.When a slow round is complete, there will be a delay while the current fast round is being completed.In this way a program can play both fast and slow games or both, but the slow games will get precedent.The fast games will be played basically to fill the time between long slow rounds and also enable a bot to get both a slow and fast rating. This won't be that complicated to add.If it works, we could do this with 9x9 games too. There are some optimizations too. If a slow round completes and none of the fast programs want to play slow games, we can start the next slow round immediately. Basically, the scheduling is exactly the same as I do now, there are just two sets of scheduling rounds, one fast and one slow. The only other difference is that when a slow round completes, the server waits on any slow players who might be playing a fast game. I think I would make the fast games significantly faster than the slow games, so that the wait between slow rounds is minimal. But that of course would be configurable. You will be able to specify that you only want fast games if that's what you want. Or that you only want slow games. The default will be both types of games. In this way, we can get the best of both worlds. Fast bots can play fast games exclusively if that's what they want to do. The variety of opponents will be a little more limited for fast bots, but as long as there are at least 2 bots willing to play a fast game, a game will happen. - Don [EMAIL PROTECTED] wrote: Couldn't there just be two servers? There were multiple volunteers. A server with long games might draw more viewers but fewer participants. Shorter games would be more helpful for those of us working on weak 19x19 programs that other people are less interested in anyway. - Dave Hillis -Original Message- From: terry mcintyre [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Sun, 28 Oct 2007 2:16 pm Subject: Re: [computer-go] 19x19 CGOS Regarding Don Dailiey's rationale for CGOS and 30-minute (or longer) time controls: a hearty AMEN! The goal here is to improve the quality of play - not merely at blitz pace, but at slower rate more comparable to the pace of humans. Some older programs peak at 10 minutes for a 19x19 game; they were designed to run on 50 MHz machines, a decade back. It might be that, for the short term, variations of Monte Carlo on quad cpus can make better use of 30 minute or longer time controls than the traditional single-threaded programs. What better incentive to the developers to try multi-threading? They'll need a strong incentive to do so, since it is a non-trivial step. But consumers of Go programs will benefit from stronger, more interesting competition. Don's idea of packing in blitz games between the longer games makes a lot of sense; it would enable a second track for those who want results more quickly. Many thanks to Don and everyone else for making CGOS possible! Terry McIntyre [EMAIL PROTECTED] mailto:[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 __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com http://mail.yahoo.com/ ___ computer-go mailing list computer-go@computer-go.org mailto:computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ *Check Out the new free AIM(R) Mail* http://o.aolcdn.com/cdn.webmail.aol.com/mailtour/aim/en-us/index.htm -- Unlimited 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] 19x19 CGOS
David Fotland wrote: It's hard to believe crazy stone is 7 stones stronger than mfgo. I'd like to see some handicap games to show this. 100 ELO might have some relation 1 handicap stone at low ratings, but at higher strengths, 1 stone handicap must be a smaller ELO difference. David This handicap can be simulated of course. It's wouldn't be exactly the same as a real handicap game, but the program giving the handicap could agree to pass on the first N moves. However, with CGOS scoring the weaker program could immediately pass and win the game! Some of my programs would just automatically pass if it won the game on score. - Don ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
What if one program agreed to moving at a1 on the first move? Would this simulate a handicap pretty well? You could get up to 4 (or is it 5) by agreeing to move to various corner intersections. Is it better to pass than move A1 on the first move? I suggest it might be interesting if the really strong programs post versions that do this. - Don Don Dailey wrote: David Fotland wrote: It's hard to believe crazy stone is 7 stones stronger than mfgo. I'd like to see some handicap games to show this. 100 ELO might have some relation 1 handicap stone at low ratings, but at higher strengths, 1 stone handicap must be a smaller ELO difference. David This handicap can be simulated of course. It's wouldn't be exactly the same as a real handicap game, but the program giving the handicap could agree to pass on the first N moves. However, with CGOS scoring the weaker program could immediately pass and win the game! Some of my programs would just automatically pass if it won the game on score. - 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] 19x19 CGOS
On Sun, 2007-10-28 at 17:05 -0400, Don Dailey wrote: Hi Dave, Two servers is easy, but 1 server is better.The plan is that I will combine fast and slow games into one server.When a slow round is complete, there will be a delay while the current fast round is being completed.In this way a program can play both fast and slow games or both, but the slow games will get precedent.The fast games will be played basically to fill the time between long slow rounds and also enable a bot to get both a slow and fast rating. This sounds like a reasonable compromise. I have only one question: What about slow players that want fast games too? I assume most slow players will play slow games and therefore use up nearly all of the time available. Depending on the pool of slow players, some dual fast+slow players may only play slow games. I have two ideas that might help solve this if it's an issue: 1. Occasionally allow a fast round between slow rounds 2. Occasionally have a dual player sit out from a slow round if the number of slow games to fast games exceeds some threshold. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
gtp has specific support for handicap games. If we do handicap, I'd prefer to see the server use those specialized commands. On Sun, 2007-10-28 at 17:21 -0400, Don Dailey wrote: What if one program agreed to moving at a1 on the first move? Would this simulate a handicap pretty well? You could get up to 4 (or is it 5) by agreeing to move to various corner intersections. Is it better to pass than move A1 on the first move? I suggest it might be interesting if the really strong programs post versions that do this. - Don Don Dailey wrote: David Fotland wrote: It's hard to believe crazy stone is 7 stones stronger than mfgo. I'd like to see some handicap games to show this. 100 ELO might have some relation 1 handicap stone at low ratings, but at higher strengths, 1 stone handicap must be a smaller ELO difference. David This handicap can be simulated of course. It's wouldn't be exactly the same as a real handicap game, but the program giving the handicap could agree to pass on the first N moves. However, with CGOS scoring the weaker program could immediately pass and win the game! Some of my programs would just automatically pass if it won the game on score. - 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] 19x19 CGOS (CS vs MFG)
Oops, I forgot to tell it to randomize. I'll restart it with random turned on. David -Original Message- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of Rémi Coulom Sent: Sunday, October 28, 2007 1:39 PM To: computer-go Subject: Re: [computer-go] 19x19 CGOS (CS vs MFG) Edward de Grijs wrote: The CrazyStone row has dissapeared because not enough games were played, so there will be a larger standard deviation around those values (I expect a 1 sigma value of about 50 elo. It would be interesting to incluse those numbers on every row (Don?)) Uncertainty about the rating is much more. Also I stopped CS-8-26-10k-1CPU, because it was deterministic, and so is Many Faces. So they were playing the same game again and again. I have now connected a parallel version, running on two cores, which makes it random. Rémi ___ 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] 19x19 CGOS
A lot of times there will be an odd number of players, in which case a random slow player will sit out (but would get to play fast games.) - Don Jason House wrote: On Sun, 2007-10-28 at 17:05 -0400, Don Dailey wrote: Hi Dave, Two servers is easy, but 1 server is better.The plan is that I will combine fast and slow games into one server.When a slow round is complete, there will be a delay while the current fast round is being completed.In this way a program can play both fast and slow games or both, but the slow games will get precedent.The fast games will be played basically to fill the time between long slow rounds and also enable a bot to get both a slow and fast rating. This sounds like a reasonable compromise. I have only one question: What about slow players that want fast games too? I assume most slow players will play slow games and therefore use up nearly all of the time available. Depending on the pool of slow players, some dual fast+slow players may only play slow games. I have two ideas that might help solve this if it's an issue: 1. Occasionally allow a fast round between slow rounds 2. Occasionally have a dual player sit out from a slow round if the number of slow games to fast games exceeds some threshold. ___ 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] CGOS 19 TC
Hi Don, Thanks very much for your effort to work cgoses. I'd like to support your idea and expect you will implement it. -Hideki Don Dailey: [EMAIL PROTECTED]: My thoughts are that it would fragment the players, we would get much less activity on either server and we need lot's of variety. However, I'm really liking the idea of playing quick 19x19 matches between rounds. It's the best of both worlds.I would arrange it so that this would never prevent you from playing the long games too. And you could choose to only play in the fast or slow games or both. - Don Joshua Shriver wrote: There is a lot of talk about time controls, and would like to add my input. I agree we should have longer time controls. I'm in the very early stages of my Go engine. With my current time line I dont anticipate having a running engine for at least a year. My design is a good bit different than the engines I've seen. Guess my question is, how resource intensive is the server code? Could we split it? Have the server running on one port for 10min blitz, and another port with 30+minute games? My only concern is fragmenting the amount of players. Not sure how many active people play, but if it's little it might make the pool to small. What are all of your thoughts? -Josh ___ 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/ -- [EMAIL PROTECTED] (Kato) ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On Sun, 2007-10-28 at 17:33 -0400, Don Dailey wrote: A lot of times there will be an odd number of players, in which case a random slow player will sit out (but would get to play fast games.) The odd number thing won't help two dual speed bots play each other at fast settings. Of course, neither did my option #2 :) - Don Jason House wrote: On Sun, 2007-10-28 at 17:05 -0400, Don Dailey wrote: Hi Dave, Two servers is easy, but 1 server is better.The plan is that I will combine fast and slow games into one server.When a slow round is complete, there will be a delay while the current fast round is being completed.In this way a program can play both fast and slow games or both, but the slow games will get precedent.The fast games will be played basically to fill the time between long slow rounds and also enable a bot to get both a slow and fast rating. This sounds like a reasonable compromise. I have only one question: What about slow players that want fast games too? I assume most slow players will play slow games and therefore use up nearly all of the time available. Depending on the pool of slow players, some dual fast+slow players may only play slow games. I have two ideas that might help solve this if it's an issue: 1. Occasionally allow a fast round between slow rounds 2. Occasionally have a dual player sit out from a slow round if the number of slow games to fast games exceeds some threshold. ___ 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] 19x19 CGOS
Hi Don, Sounds like a good idea. - Dave Hillis -Original Message- From: Don Dailey [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Sun, 28 Oct 2007 5:05 pm Subject: Re: [computer-go] 19x19 CGOS Hi Dave, Two servers is easy, but 1 server is better.The plan is that I will combine fast and slow games into one server.When a slow round is complete, there will be a delay while the current fast round is being completed.In this way a program can play both fast and slow games or both, but the slow games will get precedent.The fast games will be played basically to fill the time between long slow rounds and also enable a bot to get both a slow and fast rating. This won't be that complicated to add.If it works, we could do this with 9x9 games too. There are some optimizations too. If a slow round completes and none of the fast programs want to play slow games, we can start the next slow round immediately. Basically, the scheduling is exactly the same as I do now, there are just two sets of scheduling rounds, one fast and one slow. The only other difference is that when a slow round completes, the server waits on any slow players who might be playing a fast game. I think I would make the fast games significantly faster than the slow games, so that the wait between slow rounds is minimal. But that of course would be configurable. You will be able to specify that you only want fast games if that's what you want. Or that you only want slow games. The default will be both types of games. In this way, we can get the best of both worlds. Fast bots can play fast games exclusively if that's what they want to do. The variety of opponents will be a little more limited for fast bots, but as long as there are at least 2 bots willing to play a fast game, a game will happen. - Don [EMAIL PROTECTED] wrote: Couldn't there just be two servers? There were multiple volunteers. A server with long games might draw more viewers but fewer participants. Shorter games would be more helpful for those of us working on weak 19x19 programs that other people are less interested in anyway. - Dave Hillis -Original Message- From: terry mcintyre [EMAIL PROTECTED] To: computer-go computer-go@computer-go.org Sent: Sun, 28 Oct 2007 2:16 pm Subject: Re: [computer-go] 19x19 CGOS Regarding Don Dailiey's rationale for CGOS and 30-minute (or longer) time controls: a hearty AMEN! The goal here is to improve the quality of play - not merely at blitz pace, but at slower rate more comparable to the pace of humans. Some older programs peak at 10 minutes for a 19x19 game; they were designed to run on 50 MHz machines, a decade back. It might be that, for the short term, variations of Monte Carlo on quad cpus can make better use of 30 minute or longer time controls than the traditional single-threaded programs. What better incentive to the developers to try multi-threading? They'll need a strong incentive to do so, since it is a non-trivial step. But consumers of Go programs will benefit from stronger, more interesting competition. Don's idea of packing in blitz games between the longer games makes a lot of sense; it would enable a second track for those who want results more quickly. Many thanks to Don and everyone else for making CGOS possible! Terry McIntyre [EMAIL PROTECTED] mailto:[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 __ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com http://mail.yahoo.com/ ___ computer-go mailing list computer-go@computer-go.org mailto:computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ *Check Out the new free AIM(R) Mail* http://o.aolcdn.com/cdn.webmail.aol.com/mailtour/aim/en-us/index.htm -- Unlimited 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/ Check Out the new free AIM(R) Mail -- Unlimited 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] 19x19 CGOS
On 28, Oct 2007, at 7:59 AM, Edward de Grijs wrote: Subject: RE: [computer-go] 19x19 CGOS From: [EMAIL PROTECTED] On Sun, 2007-10-28 at 11:54 +0100, Edward de Grijs wrote: Hi all, For CGOS 19x19 I prefer a short time control (10min/game) because: 1) More quickly a more accurate rating can be established. I agree with Don. 10 minutes sudden death is brutally short for 19x19. You are limiting the pool and strength of programs available for CGOS. Hi, maybe so, but can you name some programs which cannot cope with 10 minutes thinking time for 19x19? SlugGo can cope with 10 minutes for a 19x19 game, but because of the way it is a wrapper over GnuGo with different and parallel search, at 10 min per game it is almost indistinguishable from Gnu. There is not enough time to search enough for SlugGo to choose different moves. Cheers, David ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
On Oct 28, 2007, at 2:37 PM, Don Dailey wrote Jason House wrote: gtp has specific support for handicap games. If we do handicap, I'd prefer to see the server use those specialized commands. Of course that's better, but I'm talking about a quick and dirty solution. I may never implement handicap games since it's tricky with ELO ratings. I suggest just making them a new instance: eg. 'MFGO-2H' == ManyFaces giving 2 stones handicap Christoph ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS handicaps
I agree that a lengthy discussion right now is probably not needed, but I want to toss in a thought: Every now and again, perhaps every 3 months, turn off ELO rating and instead start using a variant of the 3 games in a row method for a fixed period of time, perhaps 2 weeks. Many players at clubs do this: after N games in a row won or lost between a specific pair of players, change the handicap by one. On cgos it would not have to be the exact same player, but rather just another player with a similar ELO rating. Eventually we should find a rough correspondence, or a curve, between ELO difference and handicap, and that could be used as the starting point in the next handicap session. I think this is in line the tournament purpose of cgos. Cheers, David On 28, Oct 2007, at 2:37 PM, Don Dailey wrote: Jason House wrote: gtp has specific support for handicap games. If we do handicap, I'd prefer to see the server use those specialized commands. Of course that's better, but I'm talking about a quick and dirty solution. I may never implement handicap games since it's tricky with ELO ratings. We talked about this at one time and it was a very complicated issue. Do we award compensation for handicap stones, etc. Also there is an issue of how to figure this into the ELO rating formula. I don't want to get into another round of discussing this right now. I might implement this in a later server version but probably not any time soon. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS
This sounds very good to me. Cheers, David On 28, Oct 2007, at 2:05 PM, Don Dailey wrote: The plan is that I will combine fast and slow games into one server.When a slow round is complete, there will be a delay while the current fast round is being completed.In this way a program can play both fast and slow games or both, but the slow games will get precedent.The fast games will be played basically to fill the time between long slow rounds and also enable a bot to get both a slow and fast rating. ... You will be able to specify that you only want fast games if that's what you want. Or that you only want slow games. The default will be both types of games. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] 19x19 CGOS handicaps
I think I would handle this by assuming 100 ELO is 1 stone handicap. The data on CGOS would eventually tell me if this should be adjusted. Or I would probably just make it self adjusting. - Don David Doshay wrote: I agree that a lengthy discussion right now is probably not needed, but I want to toss in a thought: Every now and again, perhaps every 3 months, turn off ELO rating and instead start using a variant of the 3 games in a row method for a fixed period of time, perhaps 2 weeks. Many players at clubs do this: after N games in a row won or lost between a specific pair of players, change the handicap by one. On cgos it would not have to be the exact same player, but rather just another player with a similar ELO rating. Eventually we should find a rough correspondence, or a curve, between ELO difference and handicap, and that could be used as the starting point in the next handicap session. I think this is in line the tournament purpose of cgos. Cheers, David On 28, Oct 2007, at 2:37 PM, Don Dailey wrote: Jason House wrote: gtp has specific support for handicap games. If we do handicap, I'd prefer to see the server use those specialized commands. Of course that's better, but I'm talking about a quick and dirty solution. I may never implement handicap games since it's tricky with ELO ratings. We talked about this at one time and it was a very complicated issue. Do we award compensation for handicap stones, etc. Also there is an issue of how to figure this into the ELO rating formula. I don't want to get into another round of discussing this right now. I might implement this in a later server version but probably not any time soon. ___ 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/